Datasets:

jamescalam

/

llama-2-arxiv-papers-chunked

like 17

1511.08119

10

twofold: First, it enables efficient inference when using densely-connected pairwise potentials. Multiple works, [10,33] have shown that dense pairwise connections result in substantial accuracy improvements, particularly at image boundaries [12, 17]. Secondly, we keep all our mean field updates differentiable with respect to their inputs as well as the CRF parameters introduced. This design enables us to use backpropagation to automatically learn all the parameters in the introduced potentials. We use two types of higher order potential, one based on object detections and the other based on superpixels. These are detailed in Sections 4.1 and 4.2 respectively. Our complete CRF model is represented by E(x) =/summationdisplay iψU i(xi) +/summationdisplay i < jψP ij(xi,xj) +/summationdisplay dψDet d(xd) +/summationdisplay sψSP s(xs),(1) where the first two terms ψU i(.)andψP ij(.,.)are the usual unary and densely-connected pairwise energies [17] and the last two terms are the newly introduced higher order

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

11

i(.)andψP ij(.,.)are the usual unary and densely-connected pairwise energies [17] and the last two terms are the newly introduced higher order energies. Energies from the object detection take the form ψDet d(xd), where vector xd is formed by elements of xthat correspond to the foreground pixels of the dthobject detection. Superpixel label consistency based energies take the form ψSP s(xs), where xs is formed by elements of xthat correspond to the pixels belonging to the sthsuperpixel. 4.1 Object Detection Based Potentials Semantic segmentation errors can be classified into two broad categories [35]: recognition and boundary errors. Boundary errors occur when semantic labels are incorrect at the edges of objects, and it has been shown that densely connected CRFs with appearanceconsistency terms are effective at combating this problem [17]. On the other hand, recognition errors occur when object categories are recognised incorrectly or not at all. A CRF with only unary and pairwise potentials cannot effectively correct these errors since they are caused by poor unary classification. However, we propose that a state-of-the-art object detector [36,37] capable of recognising and localising objects, can

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

12

state-of-the-art object detector [36,37] capable of recognising and localising objects, can provide important information in this situation and help reduce the recognition error, as shown in Fig. 2. A key challenge in feeding-in object-detection potentials to semantic segmentation are false detections. A na ¨ıve approach of adding an object detector’s output to a CRF formulated to solve the problem of semantic segmentation would confuse the CRF due to the presence of the false positives in the detector’s output. Therefore, a robust formulation, which can automatically reject object detection false positives when they

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

13

Higher Order Conditional Random Fields in Deep Neural Networks 7 positive: some of the detections obtained from the object detector may be identified to be false following CRF inference. AllYdvariables are added to the CRF which previously contained only Xivariables. Let each (Xd,Yd), where {Xd}={Xi∈ {X}|i∈Fd}, form a clique cdin the CRF. We define the detection-based higher order energy associated with a particular assignment (xd,yd)to the clique (Xd,Yd)as follows: ψDet d(Xd=xd,Yd=yd) =  wDetsd nd/summationtextnd i=1[x(i) d=ld]ifyd= 0, wDetsd nd/summationtextnd i=1[x(i) d/ne}ationslash=ld]ifyd= 1,(2) wherend=|Fd|is the number of foreground pixels in the dthdetection,x(i) dis theith element of the vector xd,wDetis a learnable weight parameter, and [.]is the Iverson

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

14

dis theith element of the vector xd,wDetis a learnable weight parameter, and [.]is the Iverson bracket. Note that this potential encourages X(i) ds to take the value ldwhenYdis1, and at the same time encourages Ydto be 0when manyX(i) ds do not take ld. In other words, it enforces the consistency among X(i) ds andYd. An important property of the above definition of ψDet d(.)is that it can be simplified as a sum of pairwise potentials between Ydand eachX(i) dfori= 1,2,...,n d. That is, ψDet d(Xd=xd,Yd=yd) =nd/summationdisplay i=1fd(x(i) d,yd), where, fd(x(i) d,yd) =  wDetsd nd[x(i) d=ld] ifyd= 0, wDetsd nd[x(i) d/ne}ationslash=ld] ifyd= 1.(3)

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

15

d=ld] ifyd= 0, wDetsd nd[x(i) d/ne}ationslash=ld] ifyd= 1.(3) We make use of this simplification in Section 5 when deriving the mean field updates associated with this potential. For the latent Yvariables, in addition to the joint potentials with Xvariables, described in Eq. (2)and(3), we also include unary potentials, which are initialised from the score sdof the object detection. The underlying idea is that if the object detector detects an object with high confidence, the CRF in turn starts with a high initial confidence about the validity of that detection. This confidence can, of course, change during the CRF inference depending on other information ( e.g.segmentation unary potentials) available to the CRF. Examples of input images with multiple detections and GrabCut foreground masks are shown in Figure 3. Note how false detections are ignored and erroneous parts of the foreground mask are also largely ignored. 4.2 Superpixel Based Potentials The next type of higher order potential we use is based on the idea that superpixels

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

16

foreground mask are also largely ignored. 4.2 Superpixel Based Potentials The next type of higher order potential we use is based on the idea that superpixels obtained from oversegmentation [39, 40] quite often contain pixels from the same visual object. It is therefore natural to encourage pixels inside a superpixel to have the same semantic label. Once again, this should not be a hard constraint in order to keep the 8 Arnab et al. (a) (b) (c) (d) Fig. 3: Effects of imperfect foreground segmentation (a,b) Detected objects, as well as the foreground masks obtained from GrabCut. (c,d) Output using detection potentials. Incorrect parts of the foreground segmentation of the main aeroplane, and entire TV detection have been ignored by CRF inference as they did not agree with the other energy terms. The person is a failure case though as the detection has caused part of the sofa to be erroneously labelled. (a) (b) (c) Fig. 4: Segmentation enhancement from superpixel based potentials (a) The output of our system without any superpixel potentials. (b) Superpixels obtained from the image using the method of [39]. Only one “layer” of superpixels is shown. In practice, we

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

17

using the method of [39]. Only one “layer” of superpixels is shown. In practice, we used four. (c) The output using superpixel potentials. The result has improved as we encourage consistency over superpixel regions. This removes some of the spurious noise that was present previously. algorithm robust to initial superpixel segmentation errors and to violations of this key assumption. We use two types of energies in the CRF to encourage superpixel consistency in semantic segmentation. Firstly, we use the Pn-Potts model type energy [41], which is described by, ψSP s(Xs=xs) =  wLow(l) if allx(i) s=l, wHigh otherwise,(4) wherewLow(l)<w High for alll, and{Xs} ⊂ {X}is a clique defined by a superpixel. The primary idea is that assigning different labels to pixels in the same superpixel incurs a higher cost, whereas one obtains a lower cost if the labelling is consistent throughout the superpixel. Costs wLow(l)andwHigh are learnable during the end-to-end training of the network.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

18

the superpixel. Costs wLow(l)andwHigh are learnable during the end-to-end training of the network. Secondly, to make this potential stronger, we average initial unary potentials from the classifier (the CNN in our case), across all pixels in the superpixel and use the average Higher Order Conditional Random Fields in Deep Neural Networks 9 as an additional unary potential for those pixels. During experiments, we observed that superpixel based higher order energy helps in getting rid of small spurious regions of wrong labels in the segmentation output, as shown in Fig. 4. 5 Mean Field Updates and Their Differentials This section discusses the mean field updates for the higher order potentials previously introduced. These update operations are differentiable with respect to the Qi(Xi)distribution inputs at each iteration, as well as the parameters of our higher order potentials. This allows us to train our CRF end-to-end as another layer of a neural network. Take a CRF with random variables V1,V2,...,V Nand a set of cliques C, which includes unary, pairwise and higher order cliques. Mean field inference approximates the joint distribution Pr(V=v)with the product of marginals/producttext iQ(Vi=vi). We

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

19

the joint distribution Pr(V=v)with the product of marginals/producttext iQ(Vi=vi). We useQ(Vc=vc)to denote the marginal probability mass for a subset {Vc}of these variables. Where there is no ambiguity, we use the short-hand notation Q(vc)to represent Q(Vc=vc). General mean field updates of such a CRF take the form [13] Qt+1(Vi=v) =1 Ziexp −/summationdisplay c∈C/summationdisplay {vc|vi=v}Qt(vc−i)ψc(vc) , (5) whereQtis the marginal after the tthiteration, vcan assignment to all variables in cliquec,vc−ian assignment to all variables in cexcept forVi,ψc(vc)is the cost of assigning vcto the clique c, andZiis the normalisation constant that makes Q(Vi=v) a probability mass function after the update. Updates from Detection Based Potentials Following Eq. (3)above, we now use Eq.(5)to derive the mean field updates related to ψDet

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

20

Updates from Detection Based Potentials Following Eq. (3)above, we now use Eq.(5)to derive the mean field updates related to ψDet d. The contribution from ψDet dto the update of Q(X(i) d=l)takes the form /summationdisplay {(xd,yd)|x(i) d=l}Q(xd−i,yd)ψDet d(xd,yd) =  wDetsd ndQ(Yd= 0) ifl=ld, wDetsd ndQ(Yd= 1) otherwise,(6) where xd−iis an assignment to Xdwith theithelement deleted. Using the same equations, we derive the contribution from the energy ψDet dto the update of Q(Yd=b)to take the form /summationdisplay {(xd,yd)|yd=b}Q(xd)ψDet d(xd,yd) =  wDetsd nd/summationtextnd i=1Q(X(i) d=ld) ifb= 0, wDetsd nd/summationtextnd

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

21

nd/summationtextnd i=1Q(X(i) d=ld) ifb= 0, wDetsd nd/summationtextnd i=1(1−Q(X(i) d=ld))otherwise. (7) It is possible to increase the number of parameters in ψDet d(.). Since we use backpropagation to learn these parameters automatically during end-to-end training, it is desirable to have a high number of parameters to increase the flexibility of the model. Following 10 Arnab et al. this idea, we made the weight wDetclass specific, that is, a function wDet(ld)is used instead ofwDetin Eqs. (2),(6)and (7). The underlying assumption is that detector outputs can be very helpful for certain classes, while being not so useful for classes that the detector performs poorly on, or classes for which foreground segmentation is often inaccurate. Note that due to the presence of detection potentials in the CRF, error differentials calculated with respect to the Xvariable unary potentials and pairwise parameters will no longer be valid in the forms described in [10]. The error differentials with respect to

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

22

calculated with respect to the Xvariable unary potentials and pairwise parameters will no longer be valid in the forms described in [10]. The error differentials with respect to theXandYvariables, as well as class-specific detection potential weights wDet(l)are included in the supplementary material. Updates for Superpixel Based Potentials The contribution from the Pn-Potts type potential to the mean field update of Q(xi=l), where pixel iis in the superpixel clique s, was derived in [7] as /summationdisplay {xs|x(i) s=l}Q(xs−i)ψSP s(xs) =wLow(l)/productdisplay j∈c,j/ne}ationslash=iQ(Xj=l)+wHigh 1−/productdisplay j∈c−iQ(Xj=l) . (8) This update operation is differentiable with respect to the parameters wLow(l)andwHigh, allowing us to optimise them via backpropagation, and also with respect to the Q(X) values enabling us to optimise previous layers in the network.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

23

allowing us to optimise them via backpropagation, and also with respect to the Q(X) values enabling us to optimise previous layers in the network. Convergence of parallel mean field updates Mean field with parallel updates, as proposed in [17] for speed, does not have any convergence guarantees in the general case. However, we usually empirically observed convergence with higher order potentials, without damping the mean field update as described in [7, 42]. This may be explained by the fact that the unaries from the initial pixelwise-prediction part of our network provide a good initialisation. In cases where the mean field energy did not converge, we still empirically observed good final segmentations. 6 Experiments We evaluate our new CRF formulation on two different datasets using the CRF-RNN network [10] as the main baseline, since we are essentially enriching the CRF model of [10]. We then present ablation studies on our models. 6.1 Experimental set-up and results Our deep network consists of two conceptually different, but jointly trained stages. The first, “unary” part of our network is formed by the FCN-8s architecture [3]. It is initialised

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

24

first, “unary” part of our network is formed by the FCN-8s architecture [3]. It is initialised from the Imagenet-trained VGG-16 network [2], and then fine-tuned with data from the VOC 2012 training set [4], extra VOC annotations of [43] and the MS COCO [44] dataset. Higher Order Conditional Random Fields in Deep Neural Networks 11 Table 1: Comparison of each higher order potential with baseline on VOC 2012 reduced validation set MethodReduced val set(%) Baseline (unary + pairwise) [10] 72.9 Superpixels only 74.0 Detections only 74.9 Detections and Superpixels 75.8Table 2: Mean IoU accuracy on VOC 2012 test set. All methods are trained with MS COCO [44] data Method Test set(%) Ours 77.9 DPN [9] 77.5 Centrale Super Boundaries [45] 75.7 Dilated Convolutions [46] 75.3 BoxSup [35] 75.2 DeepLab Attention [47] 75.1 CRF-RNN (baseline) [10] 74.7 DeepLab WSSL [48] 73.9

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

25

DeepLab Attention [47] 75.1 CRF-RNN (baseline) [10] 74.7 DeepLab WSSL [48] 73.9 DeepLab [12] 72.7 Table 3: Mean Intersection over Union (IoU) results on PASCAL Context validation set compared to other current methods. Method Ours BoxSup [35] ParseNet [49] CRF-RNN [10] FCN-8s [3] CFM [28] Mean IoU (%) 41.3 40.5 40.4 39.3 37.8 34.4 The output of the first stage is fed into our CRF inference network. This is implemented using the mean field update operations and their differentials described in Section 5. Five iterations of mean field inference were performed during training. Our CRF network has two additional inputs in addition to segmentation unaries obtained from the FCN-8s network: data from the object detector and superpixel oversegmentations of the image. We used the publicly available code and model of the Faster R-CNN [37] object detector. The fully automated version of GrabCut [38] was then used to obtain foregrounds from the detection bounding boxes. These choices were made after conducting preliminary experiments with alternate detection and foreground segmentation algorithms.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

26

from the detection bounding boxes. These choices were made after conducting preliminary experiments with alternate detection and foreground segmentation algorithms. Four levels of superpixel oversegmentations were used, with increasing superpixel size to define the cliques used in this potential. Four levels were used since performance on the VOC validation set stopped increasing after this number. We used the superpixel method of [39] as it was shown to adhere to object boundaries the best [40], but our method generalises to any oversegmentation algorithm. We trained the full network end-to-end, optimising the weights of the CNN classifier (FCN-8s) and CRF parameters jointly. We initialised our network using the publicly available weights of [10], and trained with a learning rate of 10−10and momentum of 0.99. The learning rate is low because the loss was not normalised by the number of pixels in the training image. This is to have a larger loss for images with more pixels. When training our CRF, we only used VOC 2012 data [4] as it has the most accurate labelling, particularly around boundaries. 12 Arnab et al. PASCAL VOC 2012 Dataset The improvement obtained by each higher order potential was evaluated on the same reduced validation set [3] used by our baseline [10]. As Table

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

27

PASCAL VOC 2012 Dataset The improvement obtained by each higher order potential was evaluated on the same reduced validation set [3] used by our baseline [10]. As Table 1 shows, each new higher order potential improves the mean IoU over the baseline. We only report test set results for our best method since the VOC guidelines discourage the use of the test set for ablation studies. On the test set (Table 2), we outperform our baseline by 3.2% which equates to a 12.6% reduction in the error rate. This sets a new state-of-the-art on the VOC dataset. Qualitative results highlighting success and failure cases of our algorithm, as well as more detailed results, are shown in our supplementary material. PASCAL Context Table 3 shows our state-of-the-art results on the recently released PASCAL Context dataset [50]. We trained on the provided training set of 4998 images, and evaluated on the validation set of 5105 images. This dataset augments VOC with annotations for all objects in the scene. As a result, there are 59 classes as opposed to the 20 in the VOC dataset. Many of these new labels are “stuff” classes such as “grass” and “sky”. Our object detectors are therefore only trained for 20 of the 59 labels in this

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

28

and “sky”. Our object detectors are therefore only trained for 20 of the 59 labels in this dataset. Nevertheless, we improve by 0.8% over the previous state-of-the-art [35] and 2% over our baseline [10]. 6.2 Ablation Studies We perform additional experiments to determine the errors made by our system, show the benefits of end-to-end training and compare our detection potentials to a simpler baseline. Unless otherwise stated, these experiments are performed on the VOC 2012 reduced validation set. Error Analysis To analyse the improvements made by our higher order potentials, we separately evaluate the performance on the “boundary” and “interior” regions in a similar manner to [35]. As shown in Fig. 5 c) and d), we consider a narrow band (trimap [16]) around the “void” labels annotated in the VOC 2012 reduced validation set. The mean IoU of pixels lying within this band is termed the “Boundary IoU” whilst the “Interior IoU” is evaluated outside this region. Fig. 5 shows our results as the trimap width is varied. Adding the detection potentials

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

29

IoU” is evaluated outside this region. Fig. 5 shows our results as the trimap width is varied. Adding the detection potentials improves the Interior IoU over our baseline (only pairwise potentials [10]) as the object detector may recognise objects in the image which the pixelwise classification stage of our network may have missed out. However, the detection potentials also improve the Boundary IoU for all tested trimap widths as well. Improving the recognition of pixels in the interior of an object also helps with delineating the boundaries since the strength of the pairwise potentials exerted by the Qdistributions at each of the correctly-detected pixels increase. Our superpixel priors also increase the Interior IoU with respect to the baseline. Encouraging consistency over regions helps to get rid of spurious regions of wrong labels (as shown in Fig. 4). Fig. 5 suggests that most of this improvement occurs in the interior of an object. The Boundary IoU is slightly lower than the baseline, and this may Higher Order Conditional Random Fields in Deep Neural Networks 13 a) Image (b) Ground truth (c) Boundary (d) Interior 0 5 10 15 20 25 30 35 406668707274767880 Trimap WidthInterior IoU Pairwise only

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

30

0 5 10 15 20 25 30 35 406668707274767880 Trimap WidthInterior IoU Pairwise only Superpixels Detections Detections and Superpixels 0 5 10 15 20 25 30 35 40505560657075 Trimap WidthBoundary IoU Pairwise only Superpixels Detections Detections and Superpixels (e) Interior IoU (f) Boundary IoU Fig. 5: Error analysis on VOC 2012 reduced validation set The IoU is computed for boundary and interior regions for various trimap widths. An example of the Boundary and Interior regions for a sample image using a width of 9 pixels is shown in white in the top row. Black regions are ignored in the IoU calculation. be due to the fact that superpixels do not always align correctly with the edges of an object (the “boundary recall” of various superpixel methods are evaluated in [40]). We can see that the combination of detection and superpixel potentials results in a substantial improvement in our Interior IoU. This is the primary reason our overall IoU on the VOC benchmark increases with higher order potentials. Benefits of end-to-end training Table 4 shows how end-to-end training outperforms

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

31

on the VOC benchmark increases with higher order potentials. Benefits of end-to-end training Table 4 shows how end-to-end training outperforms piecewise training. We trained the CRF piecewise by freezing the weights of the unary part of the network, and only learning the CRF parameters. Our results in Table 2 used the FCN-8s [3] architecture to generate unaries. To show that our higher order potentials improve performance regardless of the underlying CNN used for producing unaries, we also perform an experiment using our reimplementation Table 4: Comparison of mean IoU (%) obtained on VOC 2012 reduced validation set from end-to-end and piecewise training Method FCN-8s DCN Unary only, fine-tuned on COCO 68.3 68.6 Pairwise CRF trained piecewise 69.5 70.7 Pairwise CRF trained end-to-end 72.9 72.5 Higher Order CRF trained piecewise 73.6 73.5 Higher Order CRF trained end-to-end 75.8 75.0 Test set performance of best model 77.9 76.9 14 Arnab et al. of the “front-end” module proposed in the Dilated Convolution Network (DCN) of [46]

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

32

14 Arnab et al. of the “front-end” module proposed in the Dilated Convolution Network (DCN) of [46] instead of FCN-8s. Table 4 shows that end-to-end training of the CRF yields considerable improvements over piecewise training. This was the case when using either FCN-8s or DCN for obtaining the initial unaries before performing CRF inference with higher order potentials. This suggests that our CRF network module can be plugged into different architectures and achieve performance improvements. Baseline for detections To evaluate the efficacy of our detection potentials, we formulate a simpler baseline since no other methods use detection information at inference time (BoxSup [35] derives ground truth for training using ground-truth bounding boxes). Our baseline is similar to CRF-RNN [10], but prior to CRF inference, we take the segmentation mask from the object detection and add a unary potential proportional to the detector’s confidence to the unary potentials for those pixels. We then perform meanfield inference (with only pairwise terms) on these “augmented” unaries. Using this

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

33

method, the mean IoU increases from 72.9% to 73.6%, which is significantly less than the 74.9% which we obtained using only our detection potentials without superpixels (Table 1). Our detection potentials perform better since our latent Ydetection variables model whether the detection hypothesis is accepted or not. Our CRF inference is able to evaluate object detection inputs in light of other potentials. Inference increases the relative score of detections which agree with the segmentation, and decreases the score of detections which do not agree with other energies in the CRF. Figures 2 b) and d) show examples of false-positive detections that have been ignored and correct detections that have been used to refine our segmentation. Our baseline, on the other hand, is far more sensitive to erroneous detections as it cannot adjust the weight given to them during inference. 7 Conclusion We presented a CRF model with two different higher order potentials to tackle the semantic segmentation problem. The first potential is based on the intuitive idea that object detection can provide useful cues for semantic segmentation. Our formulation is capable of automatically rejecting false object detections that do not agree at all with the semantic segmentation. Secondly, we used a potential that encourages superpixels to

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

34

capable of automatically rejecting false object detections that do not agree at all with the semantic segmentation. Secondly, we used a potential that encourages superpixels to have consistent labelling. These two new potentials can co-exist with the usual unary and pairwise potentials in a CRF. Importantly, we showed that efficient mean field inference is still possible in the presence of the new higher order potentials and derived the explicit forms of the mean field updates and their differentials. This enabled us to implement the new CRF model as a stack of CNN layers and to train it end-to-end in a unified deep network with a pixelwise CNN classifier. We experimentally showed that the addition of higher order potentials results in a significant increase in semantic segmentation accuracy allowing us to reach state-of-the-art performance. This work was supported by ERC grant ERC-2012-AdG 321162-HELIOS, EPSRC grant Seebibyte EP/M013774/1, EPSRC/MURI grant EP/N019474/1 and the Clarendon Fund. Higher Order Conditional Random Fields in Deep Neural Networks 15 References

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

35

Higher Order Conditional Random Fields in Deep Neural Networks 15 References 1.Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: NIPS. (2012) 1097–1105 2.Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. In: ICLR. (2015) 3.Long, J., Shelhamer, E., Darrell, T.: Fully convolutional networks for semantic segmentation. In: CVPR. (2015) 4.Everingham, M., Van Gool, L., Williams, C.K., Winn, J., Zisserman, A.: The pascal visual object classes (voc) challenge. IJCV (2010) 5.Ladicky, L., Russell, C., Kohli, P., Torr, P.H.: Associative hierarchical crfs for object class image segmentation. In: ICCV . (2009) 739–746 6.Ladick ´y, L., Sturgess, P., Alahari, K., Russell, C., Torr, P.H.S.: What, where and how many?

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

36

combining object detectors and crfs. In: ECCV . (2010) 424–437 7.Vineet, V., Warrell, J., Torr, P.H.: Filter-based mean-field inference for random fields with higher-order terms and product label-spaces. IJCV (2014) 8.Hariharan, B., Arbel ´aez, P., Girshick, R., Malik, J.: Simultaneous detection and segmentation. In: ECCV . Springer (2014) 297–312 9.Liu, Z., Li, X., Luo, P., Loy, C.C., Tang, X.: Semantic image segmentation via deep parsing network. In: ICCV . (2015) 10.Zheng, S., Jayasumana, S., Romera-Paredes, B., Vineet, V., Su, Z., Du, D., Huang, C., Torr, P.: Conditional random fields as recurrent neural networks. In: ICCV . (2015) 11.Lin, G., Shen, C., Reid, I.: Efficient piecewise training of deep structured models for semantic segmentation. In: CVPR. (2016)

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

37

segmentation. In: CVPR. (2016) 12.Chen, L.C., Papandreou, G., Kokkinos, I., Murphy, K., Yuille, A.L.: Semantic image segmentation with deep convolutional nets and fully connected crfs. ICLR (2015) 13.Koller, D., Friedman, N.: Probabilistic graphical models: principles and techniques. MIT press (2009) 14.Shotton, J., Winn, J., Rother, C., Criminisi, A.: Textonboost for image understanding: Multiclass object recognition and segmentation by jointly modeling texture, layout, and context. IJCV (2009) 15.He, X., Zemel, R.S., Carreira-Perpi ˜n´an, M. ´A.: Multiscale conditional random fields for image labeling. In: CVPR, IEEE (2004) 16.Kohli, P., Ladicky, L., Torr, P.: Robust higher order potentials for enforcing label consistency. IJCV 82(3) (2009) 302–324 17.Kr¨ahenb ¨uhl, P., Koltun, V.: Efficient inference in fully connected CRFs with Gaussian edge

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

38

17.Kr¨ahenb ¨uhl, P., Koltun, V.: Efficient inference in fully connected CRFs with Gaussian edge potentials. In: NIPS. (2011) 18.Ladicky, L., Russell, C., Kohli, P., Torr, P.H.: Graph cut based inference with co-occurrence statistics. In: ECCV . (2010) 239–253 19.Rabinovich, A., Vedaldi, A., Galleguillos, C., Wiewiora, E., Belongie, S.: Objects in context. In: ICCV . (2007) 1–8 20.Gonfaus, J.M., Boix, X., Van de Weijer, J., Bagdanov, A.D., Serrat, J., Gonzalez, J.: Harmony potentials for joint classification and segmentation. In: CVPR, IEEE (2010) 3280–3287 21.Yao, J., Fidler, S., Urtasun, R.: Describing the Scene as a Whole: Joint Object Detection, Scene Classification and Semantic Segmentation. In: CVPR. (2012) 702–709

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

39

Scene Classification and Semantic Segmentation. In: CVPR. (2012) 702–709 22.Wojek, C., Schiele, B.: A dynamic conditional random field model for joint labeling of object and scene classes. In: ECCV , Springer (2008) 733–747 23.Lin, G., Shen, C., Reid, I., van den Hengel, A.: Deeply learning the messages in message passing inference. In: NIPS. (2015) 361–369 16 Arnab et al. 24.Yang, Y., Hallman, S., Ramanan, D., Fowlkes, C.C.: Layered object models for image segmentation. PAMI (2012) 25.Sun, M., Kim, B.s., Kohli, P., Savarese, S.: Relating things and stuff via objectproperty interactions. PAMI 36(7) (2014) 1370–1383 26.Carreira, J., Caseiro, R., Batista, J., Sminchisescu, C.: Semantic segmentation with secondorder pooling. In: ECCV . (2012) 430–443 27.Farabet, C., Couprie, C., Najman, L., LeCun, Y.: Learning hierarchical features for scene

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

40

27.Farabet, C., Couprie, C., Najman, L., LeCun, Y.: Learning hierarchical features for scene labeling. PAMI (2013) 28.Dai, J., He, K., Sun, J.: Convolutional feature masking for joint object and stuff segmentation. CVPR (2015) 29.Tompson, J.J., Jain, A., LeCun, Y ., Bregler, C.: Joint training of a convolutional network and a graphical model for human pose estimation. In: NIPS. (2014) 1799–1807 30.Deng, Z., Zhai, M., Chen, L., Liu, Y., Muralidharan, S., Roshtkhari, M.J., Mori, G.: Deep structured models for group activity recognition. In: BMVC. (2015) 31.Ionescu, C., Vantzos, O., Sminchisescu, C.: Matrix backpropagation for deep networks with structured layers. In: ICCV . (2015) 2965–2973 32.Domke, J.: Learning graphical model parameters with approximate marginal inference. PAMI (2013) 33.Kr¨ahenb ¨uhl, P., Koltun, V.: Parameter learning and convergent inference for dense random

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

41

(2013) 33.Kr¨ahenb ¨uhl, P., Koltun, V.: Parameter learning and convergent inference for dense random fields. In: ICML. (2013) 34.Ross, S., Munoz, D., Hebert, M., Bagnell, J.A.: Learning message-passing inference machines for structured prediction. In: CVPR. (2011) 35.Dai, J., He, K., Sun, J.: Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation. In: ICCV . (2015) 36. Girshick, R.: Fast r-cnn. In: ICCV . (2015) 37.Ren, S., He, K., Girshick, R., Sun, J.: Faster R-CNN: Towards real-time object detection with region proposal networks. In: NIPS. (2015) 38.Rother, C., Kolmogorov, V., Blake, A.: Grabcut: Interactive foreground extraction using iterated graph cuts. ACM TOG (2004) 39.Felzenszwalb, P.F., Huttenlocher, D.P.: Efficient graph-based image segmentation. IJCV (2004)

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

42

(2004) 40.Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., Susstrunk, S.: Slic superpixels compared to state-of-the-art superpixel methods. PAMI 34(11) (2012) 2274–2282 41. Kohli, P., Kumar, M.P., Torr, P.H.: P3 & beyond: Solving energies with higher order cliques. In: CVPR. (2007) 42.Baqu, P., Bagautdinov, T., Fleuret, F., Fua, P.: Principled Parallel Mean-Field Inference for Discrete Random Fields. In: CVPR. (2016) 43.Hariharan, B., Arbel ´aez, P., Bourdev, L., Maji, S., Malik, J.: Semantic contours from inverse detectors. In: ICCV , IEEE (2011) 991–998 44.Lin, T.Y ., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll ´ar, P., Zitnick, C.L.: Microsoft coco: Common objects in context. In: ECCV . Springer (2014) 740–755

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

43

Microsoft coco: Common objects in context. In: ECCV . Springer (2014) 740–755 45.Kokkinos, I.: Pushing the boundaries of boundary detection using deep learning. In: ICLR. (2016) 46.Yu, F., Koltun, V .: Multi-scale context aggregation by dilated convolutions. In: ICLR. (2016) 47.Chen, L.C., Yang, Y ., Wang, J., Xu, W., Yuille, A.L.: Attention to scale: Scale-aware semantic image segmentation. In: CVPR. (2016) 48.Papandreou, G., Chen, L., Murphy, K., Yuille, A.L.: Weakly- and semi-supervised learning of a DCNN for semantic image segmentation. In: ICCV . (2015) 49.Liu, W., Rabinovich, A., Berg, A.C.: Parsenet: Looking wider to see better. arXiv preprint arXiv:1506.04579 (2015) Higher Order Conditional Random Fields in Deep Neural Networks 17 50.Mottaghi, R., Chen, X., Liu, X., Cho, N.G., Lee, S.W., Fidler, S., Urtasun, R., et al.: The

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

44

role of context for object detection and semantic segmentation in the wild. In: CVPR, IEEE (2014) 891–898 51.Noh, H., Hong, S., Han, B.: Learning deconvolution network for semantic segmentation. In: ICCV . (2015) 52.Mostajabi, M., Yadollahpour, P., Shakhnarovich, G.: Feedforward semantic segmentation with zoom-out features. In: CVPR. (2015) 53.Dong, J., Chen, Q., Yan, S., Yuille, A.: Towards unified object detection and semantic segmentation. In: ECCV . (2014) 299–314 54.Carreira, J., Caseiro, R., Batista, J., Sminchisescu, C.: Free-form region description with second-order pooling. PAMI (2014) Appendix Appendix A of this supplementary material presents the derivatives of the mean field updates which we use for inference in our Conditional Random Field (CRF). Appendix B shows detailed qualitative results for the experiments described in our main paper. A Derivatives of Mean Field Updates The pseudocode for the mean field inference algorithm with latent Ydetection variables

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

45

A Derivatives of Mean Field Updates The pseudocode for the mean field inference algorithm with latent Ydetection variables is shown below in Algorithm 1. We use the same notation used in the main paper. Algorithm 1 Mean Field Inference Q0(Xi=l)←1 Ziexp/parenleftbig −ψU i(l)/parenrightbig ,∀i,l Q0(Yd=b)←sb d(1−sd)(1−b),∀d,b ⊿Initialisation fort = 0 :T−1do Et(Xi=l)←UnaryUpdate + PairwiseUpdate + DetectionUpdate + SuperpixelUpdate, ∀i,l Et(Yd=b)←YUnaryUpdate + Y DetectionUpdate ⊿Mean field updates Qt+1(Xi=l)←1 Ziexp/parenleftbig −Et(Xi=l)/parenrightbig ,∀i,l Qt+1(Yd=b)←1 Zdexp/parenleftbig −Et(Yd=b)/parenrightbig ,∀d,b ⊿Normalising end for

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

46

Zdexp/parenleftbig −Et(Yd=b)/parenrightbig ,∀d,b ⊿Normalising end for For the explicit forms of the UnaryUpdate andPairwiseUpdate above, and their differentials, we refer the reader to [10] and discuss the terms DetectionUpdate and SuperpixelUpdate in detail below. Let us assume that only one object detection of the form (ld,sd,Fd)is available for the image under consideration. When multiple detections are present, simply a summation of the updates and differentials discussed below apply. Therefore, no generality 18 Arnab et al. is lost with this assumption. Similarly, we can assume that only one superpixel clique {Xs}is present, without a loss of generality. Assuming that pixel iin Algorithm 1 belongs to Fd, Eq. (6) in the main paper described the exact form of DetectionUpdate . Similarly, assuming that pixel ibelongs to{Xs}Eq. (8) described the form of SuperpixelUpdate . LetLdenote the value of the loss function calculated at the output of the deep network. This could be the softmax loss or any other appropriate loss function. During backpropagation, we get the error signal∂L

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

47

network. This could be the softmax loss or any other appropriate loss function. During backpropagation, we get the error signal∂L ∂QTat the output of the mean field inference. Using this error information, we need to compute the derivative of the loss Lwith respect to theXunaries and various CRF parameters. Note that, if we compute the relevent differentials for only one iteration of the mean field algorithm, it is possible to calculate them for multiple iterations using the recurrent behaviour of the iterations. Note that, by looking at Normalising step of Algorithm 1, it is trivial to calculate ∂Qt+1 ∂Et. Therefore, we can then calculate∂L ∂Etusing the chain rule. This is same as backpropagation of the usual softmax operation in a deep network (up to a negative sign). Using this observation we can calculate the necessary differentials to take the forms shown below: ∂L ∂wDet=sd ndnd/summationdisplay i=1/parenleftBig∂L Et(X(i) d=ld)Qt(Yd= 1) + (9) /summationdisplay l/prime/negationslash=ld∂L

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

48

d=ld)Qt(Yd= 1) + (9) /summationdisplay l/prime/negationslash=ld∂L ∂Et(X(i) d=l/prime)Qt(Yd= 1)/parenrightBig + ∂L ∂Et(Yd= 0)sd ndnd/summationdisplay i=1Qt(X(i) d=ld) + ∂L ∂Et(Yd= 1)sd ndnd/summationdisplay i=1/parenleftBig 1−Qt(X(i) d=ld)/parenrightBig ∂L ∂Qt(X(i) d=ld)=wDet∂L ∂Et(Yd= 0)−wDet∂L ∂Et(Yd= 1)(10) ∂L ∂Qt(Yd= 0)=wDetsd ndnd/summationdisplay i=1/parenleftBigg ∂L Et(X(i) d=ld)/parenrightBigg (11) ∂L ∂Qt(Yd= 1)=wDetsd ndnd/summationdisplay

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

49

d=ld)/parenrightBigg (11) ∂L ∂Qt(Yd= 1)=wDetsd ndnd/summationdisplay i=1/summationdisplay l/negationslash=ld/parenleftBigg ∂L ∂Et(X(i) d=l/prime)/parenrightBigg (12) ∂L ∂wLow(l)=/summationdisplay i∈s ∂L ∂Et(X(i) s=l)/productdisplay j∈c,j/negationslash=iQt(Xj=l)  (13) Higher Order Conditional Random Fields in Deep Neural Networks 19 ∂L ∂wHigh=/summationdisplay i∈s/summationdisplay l∈L ∂L ∂Et(X(i) s=l) 1−/productdisplay j∈c,j/negationslash=iQt(Xj=l)   (14) Effect of the superpixel potentials on the derivatives∂L

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

50

  (14) Effect of the superpixel potentials on the derivatives∂L ∂Qt(Xi=l)were negligible. Therefore, we ignored them in our calculations. B Additional Experimental Results Table 5 presents more detailed results of our method, and that of other state-of-the-art techniques, on the PASCAL VOC 2012 test set. In particular, we present the accuracy for every class in the VOC test set. Note that our per-class accuracy improves over our baseline, CRF-RNN [10], for all of the 20 classes in PASCAL VOC. Figure 6 shows more sample results of our system, compared to our baseline, CRFas-RNN [10]. Figure 7 shows examples of failure cases of our method. Figure 8 examines the effect of each of our potentials. Finally, Figure 9 shows a qualitative comparison between the output of our system and other current methods on the PASCAL VOC 2012 test set. 20 Arnab et al.Table 5: Comparison of the mean Intersection over Union (IoU) accuracy of our approach and other state-of-the-art methods on the Pascal VOC 2012 test set. Scores for other methods were taken from the original authors’ publications. Methods trained with COCOMean

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

51

VOC 2012 test set. Scores for other methods were taken from the original authors’ publications. Methods trained with COCOMean IoU(%)aeroplanebike bird boat bottle bus car cat chair cow table dog horse mbikepersonplant sheep sofa train tv Our method 77.9 92.5 59.1 90.3 70.6 74.4 92.4 84.1 88.3 36.8 85.6 67.1 85.1 86.9 88.2 82.6 62.6 85.0 56.2 81.9 72.5 DPN [9] 77.5 89.0 61.6 87.7 66.8 74.7 91.2 84.3 87.6 36.5 86.3 66.1 84.4 87.8 85.6 85.4 63.6 87.3 61.3 79.4 66.4 Super Bound. [45] 75.7 90.3 37.9 89.6 67.8 74.6 89.3 84.1 89.1 35.8 83.6 66.2 82.9 81.7 85.6 84.6 60.3 84.8 60.7 78.3 68.3

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

52

Dilated Conv. [46] 75.3 91.7 39.6 87.8 63.1 71.8 89.7 82.9 89.8 37.2 84.0 63.0 83.3 89.0 83.8 85.1 56.8 87.6 56.0 80.2 64.7 BoxSup [35] 75.2 89.8 38.0 89.2 68.9 68.0 89.6 83.0 87.7 34.4 83.6 67.1 81.5 83.7 85.2 83.5 58.6 84.9 55.8 81.2 70.7 Attention [47] 75.1 92.0 41.2 87.8 57.2 72.7 92.8 85.9 90.5 30.5 78.0 62.8 85.8 85.3 87.2 85.6 57.7 85.1 56.5 83.0 65.0 CRF-as-RNN [10] 74.7 90.4 55.3 88.7 68.4 69.8 88.3 82.4 85.1 32.6 78.5 64.4 79.6 81.9 86.4 81.8 58.6 82.4 53.5 77.4 70.1

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

53

WSSL [48] 73.9 89.2 46.7 88.5 63.5 68.4 87.0 81.2 86.3 32.6 80.7 62.4 81.0 81.3 84.3 82.1 56.2 84.6 58.2 76.2 67.2 DeepLab [12] 72.7 89.1 38.3 88.1 63.3 69.7 87.1 83.1 85.0 29.3 76.5 56.5 79.8 77.9 85.8 82.4 57.4 84.3 54.9 80.5 64.1 Methods trained without COCO Our method 73.9 89.3 40.0 81.6 65.1 71.7 90.1 81.3 85.7 32.4 82.1 62.2 82.6 83.7 84.5 81.1 60.8 85.2 49.6 80.0 69.9 DPN [9] 74.1 87.7 59.4 78.4 64.9 70.3 89.3 83.5 86.1 31.7 79.9 62.6 81.9 80.0 83.5 82.3 60.5 83.2 53.4 77.9 65.0

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

54

DeconvNet [51] 72.5 89.9 39.3 79.7 63.9 68.2 87.4 81.2 86.1 28.5 77.0 62.0 79.0 80.3 83.6 80.2 58.8 83.4 54.3 80.7 65.0 CRF-as-RNN [10] 72.0 87.5 39.0 79.7 64.2 68.3 87.6 80.8 84.4 30.4 78.2 60.4 80.5 77.8 83.1 80.6 59.5 82.8 47.8 78.3 67.1 DeepLab [12] 71.6 84.4 54.5 81.5 63.6 65.9 85.1 79.1 83.4 30.7 74.1 59.8 79.0 76.1 83.2 80.8 59.7 82.2 50.4 73.1 63.7 Piecewise [11] 70.7 87.5 37.7 75.8 57.4 72.3 88.4 82.6 80.0 33.4 71.5 55.0 79.3 78.4 81.3 82.7 56.1 79.8 48.6 77.1 66.3

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

55

Zoomout [52] 69.6 85.6 37.3 83.2 62.5 66.0 85.1 80.7 84.9 27.2 73.2 57.5 78.1 79.2 81.1 77.1 53.6 74.0 49.2 71.7 63.3 FCN-8s [3] 62.2 76.8 34.2 68.9 49.4 60.3 75.3 74.7 77.6 21.4 62.5 46.8 71.8 63.9 76.5 73.9 45.2 72.4 37.4 70.9 55.1 CFM [28] 61.8 75.7 26.7 69.5 48.8 65.6 81.0 69.2 73.3 30.0 68.7 51.5 69.1 68.1 71.7 67.5 50.4 66.5 44.4 58.9 53.5 NUS UDS [53] 50.0 67.0 24.5 47.2 45.0 47.9 65.3 60.6 58.5 15.5 50.8 37.4 45.8 59.9 62.0 52.7 40.8 48.2 36.8 53.1 45.6

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

56

O2P [54] 47.8 64.0 27.3 54.1 39.2 48.7 56.6 57.7 52.5 14.2 54.8 29.6 42.2 58.0 54.8 50.2 36.6 58.6 31.6 48.4 38.6 Higher Order Conditional Random Fields in Deep Neural Networks 21 Input image CRF-as-RNN [10] Our method Ground truth Fig. 6: Examples of images where our method has improved over our baseline, CRF-as-RNN [10]. The input images have the detection bounding boxes overlaid on them. Note that the method of [10] does not make use of this information. The improvements from our method are due to our detection potentials, as well as our superpixel based potentials. Note that all images are from the reduced validation set of VOC 2012 and have not been trained on at all. Best viewed in color. 22 Arnab et al. Input image CRF-as-RNN [10] Our method Ground truth Fig. 7: Examples of failure cases where our method has performed poorly. The first row shows an example of how the detection of the person has now resulted in the sofa being misclassified (although our system is able to reject the other false detection). Our

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

57

row shows an example of how the detection of the person has now resulted in the sofa being misclassified (although our system is able to reject the other false detection). Our superpixel potentials have a tendency to remove spurious noise by enforcing consistency within regions. However, as shown in the second row, sometimes the “noise” being removed is actually the correct label. In the other cases, we are limited by our pixelwise classification unaries which are poor. Our superpixel and detection potentials are not always able to compensate for this. Note that all images are from the reduced validation set of VOC 2012 and have not been trained on at all. The input images have the detection bounding boxes overlaid on them. Note that the method of [10] does not make use of this information. Best viewed in color. Higher Order Conditional Random Fields in Deep Neural Networks 23 Input image Pairwise onlySuperpixels onlyDetections onlyDetections and SuperpixelsGround truth Fig. 8: Comparison of pairwise potentials, superpixel and pairwise potentials, detection and pairwise potentials, and a combination of all three (Row 1 and 2) These are examples where superpixel potentials help to remove spurious noise in the output

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

58

are examples where superpixel potentials help to remove spurious noise in the output but detection potentials do not affect the result. The final result still improves when all potentials are combined. (Row 3) Detection potentials greatly improve the result by recognising the train correctly (the pixelwise unaries are largest for “bus”). And superpixels, when combined with detections, slightly improve the output. (Row 4) An example where both superpixel and detection potentials improve the final output. (Row 5) A case where the superpixel worsens the result as, although the output is more consistent among superpixel regions, some pixels have had their correct labels removed. However, the correct detection improves the result, and the output of combining superpixel and detection potentials is actually better than either potential in isolation. (Row 6) Here, the detection (although correct) worsens the output due to its imprecise foreground mask. Superpixel potentials also exacerbate the result, since the legs of the chair and the chair’s shadow are confused to be part of the same superpixel region. However, when the two potentials are combined, the result is slightly better than with only detection potentials. 24 Arnab et al.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

59

potentials are combined, the result is slightly better than with only detection potentials. 24 Arnab et al. Input image FCN-8s [3] Deeplab [12] CRF-as-RNN [10] Our method Ground truth Fig. 9: Qualitative comparison with other current methods. Sample results of our method compared to other current techniques on VOC 2012. We reproduced the segmentation results of Deeplab from their original publication, whilst we reproduced the results of FCN-8s and CRF-as-RNN from their publicly-available source code. Best viewed in colour.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1512.02595

0

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin Baidu Research – Silicon Valley AI Lab Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu Abstract We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech—two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

1

our application of HPC techniques, resulting in a 7x speedup over our previous system [26]. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale. 1 Introduction Decades worth of hand-engineered domain knowledge has gone into current state-of-the-art automatic speech recognition (ASR) pipelines. A simple but powerful alternative solution is to train such ASR models end-to-end, using deep learning to replace most modules with a single model [26]. We present the second generation of our speech system that exemplifies the major advantages of endto-end learning. The Deep Speech 2 ASR pipeline approaches or exceeds the accuracy of Amazon Mechanical Turk human workers on several benchmarks, works in multiple languages with little modification, and is deployable in a production setting. It thus represents a significant step towards

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

2

modification, and is deployable in a production setting. It thus represents a significant step towards a single ASR system that addresses the entire range of speech recognition contexts handled by humans. Since our system is built on end-to-end deep learning, we can employ a spectrum of deep learning techniques: capturing large training sets, training larger models with high performance computing, and methodically exploring the space of neural network architectures. We show that through these techniques we are able to reduce error rates of our previous end-to-end system [26] in English by up to 43%, and can also recognize Mandarin speech with high accuracy. One of the challenges of speech recognition is the wide range of variability in speech and acoustics. As a result, modern ASR pipelines are made up of numerous components including complex feature extraction, acoustic models, language and pronunciation models, speaker adaptation, etc. Building and tuning these individual components makes developing a new speech recognizer very hard, especially for a new language. Indeed, many parts do not generalize well across environments or languages and it is often necessary to support multiple application-specific systems in order to provide acceptable accuracy. This state of affairs is different from human speech recognition: people Authorship order is alphabetical.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

3

Authorship order is alphabetical. 1arXiv:1512.02595v1 [cs.CL] 8 Dec 2015 have the innate ability to learn any language during childhood, using general skills to learn language. After learning to read and write, most humans can transcribe speech with robustness to variation in environment, speaker accent and noise, without additional training for the transcription task. To meet the expectations of speech recognition users, we believe that a single engine must learn to be similarly competent; able to handle most applications with only minor modifications and able to learn new languages from scratch without dramatic changes. Our end-to-end system puts this goal within reach, allowing us to approach or exceed the performance of human workers on several tests in two very different languages: Mandarin and English. Since Deep Speech 2 (DS2) is an end-to-end deep learning system, we can achieve performance gains by focusing on three crucial components: the model architecture, large labeled training datasets, and computational scale. This approach has also yielded great advances in other application areas such as computer vision and natural language. This paper details our contribution to these three areas for speech recognition, including an extensive investigation of model architectures

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

4

these three areas for speech recognition, including an extensive investigation of model architectures and the effect of data and model size on recognition performance. In particular, we describe numerous experiments with neural networks trained with the Connectionist Temporal Classification (CTC) loss function [22] to predict speech transcriptions from audio. We consider networks composed of many layers of recurrent connections, convolutional filters, and nonlinearities, as well as the impact of a specific instance of Batch Normalization [63] (BatchNorm) applied to RNNs. We not only find networks that produce much better predictions than those in previous work [26], but also find instances of recurrent models that can be deployed in a production setting with no significant loss in accuracy. Beyond the search for better model architecture, deep learning systems benefit greatly from large quantities of training data. We detail our data capturing pipeline that has enabled us to create larger datasets than what is typically used to train speech recognition systems. Our English speech system is trained on 11,940 hours of speech, while the Mandarin system is trained on 9,400 hours. We use data synthesis to further augment the data during training. Training on large quantities of data usually requires the use of larger models. Indeed, our models

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

5

data synthesis to further augment the data during training. Training on large quantities of data usually requires the use of larger models. Indeed, our models have many more parameters than those used in our previous system. Training a single model at these scales requires tens of exaFLOPs1that would require 3-6 weeks to execute on a single GPU. This makes model exploration a very time consuming exercise, so we have built a highly optimized training system that uses 8 or 16 GPUs to train one model. In contrast to previous large-scale training approaches that use parameter servers and asynchronous updates [18, 10], we use synchronous SGD, which is easier to debug while testing new ideas, and also converges faster for the same degree of data parallelism. To make the entire system efficient, we describe optimizations for a single GPU as well as improvements to scalability for multiple GPUs. We employ optimization techniques typically found in High Performance Computing to improve scalability. These optimizations include a fast implementation of the CTC loss function on the GPU, and a custom memory allocator. We also use carefully integrated compute nodes and a custom implementation of all-reduce to accelerate inter-GPU communication. Overall the system sustains approximately 50 teraFLOP/second when training on 16 GPUs. This amounts to 3 teraFLOP/second per GPU which is about 50% of peak

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

6

training on 16 GPUs. This amounts to 3 teraFLOP/second per GPU which is about 50% of peak theoretical performance. This scalability and efficiency cuts training times down to 3 to 5 days, allowing us to iterate more quickly on our models and datasets. We benchmark our system on several publicly available test sets and compare the results to our previous end-to-end system [26]. Our goal is to eventually reach human-level performance not only on specific benchmarks, where it is possible to improve through dataset-specific tuning, but on a range of benchmarks that reflects a diverse set of scenarios. To that end, we have also measured the performance of human workers on each benchmark for comparison. We find that our system outperforms humans in some commonly-studied benchmarks and has significantly closed the gap in much harder cases. In addition to public benchmarks, we show the performance of our Mandarin system on internal datasets that reflect real-world product scenarios. Deep learning systems can be challenging to deploy at scale. Large neural networks are computationally expensive to evaluate for each user utterance, and some network architectures are more

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

7

Deep learning systems can be challenging to deploy at scale. Large neural networks are computationally expensive to evaluate for each user utterance, and some network architectures are more easily deployed than others. Through model exploration, we find high-accuracy, deployable network architectures, which we detail here. We also employ a batching scheme suitable for GPU 11 exaFLOP = 1018FLoating-point OPerations. 2 hardware called Batch Dispatch that leads to an efficient, real-time implementation of our Mandarin engine on production servers. Our implementation achieves a 98th percentile compute latency of 67 milliseconds, while the server is loaded with 10 simultaneous audio streams. The remainder of the paper is as follows. We begin with a review of related work in deep learning, end-to-end speech recognition, and scalability in Section 2. Section 3 describes the architectural and algorithmic improvements to the model and Section 4 explains how to efficiently compute them. We discuss the training data and steps taken to further augment the training set in Section 5. An analysis of results for the DS2 system in English and Mandarin is presented in Section 6. We end with a description of the steps needed to deploy DS2 to real users in Section 7. 2 Related Work

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

8

description of the steps needed to deploy DS2 to real users in Section 7. 2 Related Work This work is inspired by previous work in both deep learning and speech recognition. Feed-forward neural network acoustic models were explored more than 20 years ago [7, 50, 19]. Recurrent neural networks and networks with convolution were also used in speech recognition around the same time [51, 67]. More recently DNNs have become a fixture in the ASR pipeline with almost all state of the art speech work containing some form of deep neural network [42, 29, 17, 16, 43, 58]. Convolutional networks have also been found beneficial for acoustic models [1, 53]. Recurrent neural networks, typically LSTMs, are just beginning to be deployed in state-of-the art recognizers [24, 25, 55] and work well together with convolutional layers for the feature extraction [52]. Models with both bidirectional [24] and unidirectional recurrence have been explored as well. End-to-end speech recognition is an active area of research, showing compelling results when used to re-score the outputs of a DNN-HMM [23] and standalone [26]. Two methods are currently used to

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

9

to re-score the outputs of a DNN-HMM [23] and standalone [26]. Two methods are currently used to map variable length audio sequences directly to variable length transcriptions. The RNN encoderdecoder paradigm uses an encoder RNN to map the input to a fixed length vector and a decoder network to expand the fixed length vector into a sequence of output predictions [11, 62]. Adding an attentional mechanism to the decoder greatly improves performance of the system, particularly with long inputs or outputs [2]. In speech, the RNN encoder-decoder with attention performs well both in predicting phonemes [12] or graphemes [3, 8]. The other commonly used technique for mapping variable length audio input to variable length output is the CTC loss function [22] coupled with an RNN to model temporal information. The CTCRNN model performs well in end-to-end speech recognition with grapheme outputs [23, 27, 26, 40]. The CTC-RNN model has also been shown to work well in predicting phonemes [41, 54], though a lexicon is still needed in this case. Furthermore it has been necessary to pre-train the CTC-RNN network with a DNN cross-entropy network that is fed frame-wise alignments from a GMM-HMM

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

10

network with a DNN cross-entropy network that is fed frame-wise alignments from a GMM-HMM system [54]. In contrast, we train the CTC-RNN networks from scratch without the need of framewise alignments for pre-training. Exploiting scale in deep learning has been central to the success of the field thus far [36, 38]. Training on a single GPU resulted in substantial performance gains [49], which were subsequently scaled linearly to two [36] or more GPUs [15]. We take advantage of work in increasing individual GPU efficiency for low-level deep learning primitives [9]. We build on the past work in using modelparallelism [15], data-parallelism [18] or a combination of the two [64, 26] to create a fast and highly scalable system for training deep RNNs in speech recognition. Data has also been central to the success of end-to-end speech recognition, with over 7000 hours of labeled speech used in Deep Speech 1 (DS1) [26]. Data augmentation has been highly effective in improving the performance of deep learning in computer vision [39, 56, 14]. This has also been shown to improve speech systems [21, 26]. Techniques used for data augmentation in speech range

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

11

shown to improve speech systems [21, 26]. Techniques used for data augmentation in speech range from simple noise addition [26] to complex perturbations such as simulating changes to the vocal tract length and rate of speech of the speaker [31, 35]. Existing speech systems can also be used to bootstrap new data collection. In one approach, the authors use one speech engine to align and filter a thousand hours of read speech [46]. In another approach, a heavy-weight offline speech recognizer is used to generate transcriptions for tens of thousands of hours of speech [33]. This is then passed through a filter and used to re-train the recognizer, resulting in significant performance gains. We draw inspiration from these past approaches in 3 bootstrapping larger datasets and data augmentation to increase the effective amount of labeled data for our system. 3 Model Architecture A simple multi-layer model with a single recurrent layer cannot exploit thousands of hours of labelled speech. In order to learn from datasets this large, we increase the model capacity via depth. We explore architectures with up to 11 layers including many bidirectional recurrent layers and convolutional layers. These models have nearly 8 times the amount of computation per data example as the models in Deep Speech 1 making fast optimization and computation critical. In order to optimize

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

12

the models in Deep Speech 1 making fast optimization and computation critical. In order to optimize these models successfully, we use Batch Normalization for RNNs and a novel optimization curriculum we call SortaGrad. We also exploit long strides between RNN inputs to reduce computation per example by a factor of 3. This is helpful for both training and evaluation, though requires some modifications in order to work well with CTC. Finally, though many of our research results make use of bidirectional recurrent layers, we find that excellent models exist using only unidirectional recurrent layers—a feature that makes such models much easier to deploy. Taken together these features allow us to tractably optimize deep RNNs and improve performance by more than 40% in both English and Mandarin error rates over the smaller baseline models. 3.1 Preliminaries Figure 1 shows the architecture of the DS2 system which at its core is similar to the previous DS1 system [26]: a recurrent neural network (RNN) trained to ingest speech spectrograms and generate text transcriptions. Let a single utterance x(i)and label y(i)be sampled from a training set X =

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

13

text transcriptions. Let a single utterance x(i)and label y(i)be sampled from a training set X = f(x(1),y(1)), (x(2),y(2)),:::g. Each utterance, x(i), is a time-series of length T(i)where every time-slice is a vector of audio features, x(i) t,t= 0,:::,T(i)1. We use a spectrogram of power normalized audio clips as the features to the system, so x(i) t,pdenotes the power of the p’th frequency bin in the audio frame at time t. The goal of the RNN is to convert an input sequence x(i)into a final transcription y(i). For notational convenience, we drop the superscripts and use xto denote a chosen utterance and ythe corresponding label. The outputs of the network are the graphemes of each language. At each output time-step t, the RNN makes a prediction over characters, p(`tjx), where`tis either a character in the alphabet or the blank symbol. In English we have `t2fa, b, c,:::,z,space ,apostrophe ,blankg, where we have added

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

14

symbol. In English we have `t2fa, b, c,:::,z,space ,apostrophe ,blankg, where we have added theapostrophe as well as a space symbol to denote word boundaries. For the Mandarin system the network outputs simplified Chinese characters. We describe this in more detail in Section 3.9. The RNN model is composed of several layers of hidden units. The architectures we experiment with consist of one or more convolutional layers, followed by one or more recurrent layers, followed by one or more fully connected layers. The hidden representation at layer lis given byhlwith the convention that h0represents the input x. The bottom of the network is one or more convolutions over the time dimension of the input. For a context window of size c, thei-th activation at time-step tof the convolutional layer is given by hl t,i=f(wl ihl1 tc:t+c) (1) wheredenotes the element-wise product between the i-th filter and the context window of the

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

15

tc:t+c) (1) wheredenotes the element-wise product between the i-th filter and the context window of the previous layers activations, and fdenotes a unary nonlinear function. We use the clipped rectifiedlinear (ReLU) function (x) = minfmaxfx, 0g, 20gas our nonlinearity. In some layers, usually the first, we sub-sample by striding the convolution by sframes. The goal is to shorten the number of time-steps for the recurrent layers above. Following the convolutional layers are one or more bidirectional recurrent layers [57]. The forward in time !hland backward in time hlrecurrent layer activations are computed as !hl t=g(hl1 t, !hl t1) hl t=g(hl1 t, hl t+1)(2) 4 CTC SpectrogramRecurrentorGRU(Bidirectional)1D or 2DInvariantConvolutionFully Connected BatchNormalizationFigure 1: Architecture of the DS2 system used to train on both English and Mandarin speech. We explore

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

16

BatchNormalizationFigure 1: Architecture of the DS2 system used to train on both English and Mandarin speech. We explore variants of this architecture by varying the number of convolutional layers from 1 to 3 and the number of recurrent or GRU layers from 1 to 7. The two sets of activations are summed to form the output activations for the layer hl= !hl+ hl. The function g(
)can be the standard recurrent operation !hl t=f(Wlhl1 t+ !Ul !hl t1+bl) (3) whereWlis the input-hidden weight matrix, !Ulis the recurrent weight matrix and blis a bias term. In this case the input-hidden weights are shared for both directions of the recurrence. The function g(
)can also represent more complex recurrence operations such as the Long Short-Term Memory (LSTM) units [30] and the gated recurrent units (GRU) [11]. After the bidirectional recurrent layers we apply one or more fully connected layers with hl t=f(Wlhl1 t+bl) (4) The output layer Lis a softmax computing a probability distribution over characters given by p(`t=kjx) =exp(wL k
hL1

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

17

The output layer Lis a softmax computing a probability distribution over characters given by p(`t=kjx) =exp(wL k
hL1 t)P jexp(wL j
hL1 t)(5) The model is trained using the CTC loss function [22]. Given an input-output pair (x,y)and the current parameters of the network , we compute the loss function L(x,y;)and its derivative with respect to the parameters of the network rL(x,y;). This derivative is then used to update the network parameters through the backpropagation through time algorithm. In the following subsections we describe the architectural and algorithmic improvements made relative to DS1 [26]. Unless otherwise stated these improvements are language agnostic. We report results on an English speaker held out development set which is an internal dataset containing 2048 utterances of primarily read speech. All models are trained on datasets described in Section 5. We report Word Error Rate (WER) for the English system and Character Error Rate (CER) for the Mandarin system. In both cases we integrate a language model in a beam search decoding step as described in Section 3.8. 5 Architecture Hidden Units Train Dev Baseline BatchNorm Baseline BatchNorm

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

18

described in Section 3.8. 5 Architecture Hidden Units Train Dev Baseline BatchNorm Baseline BatchNorm 1 RNN, 5 total 2400 10.55 11.99 13.55 14.40 3 RNN, 5 total 1880 9.55 8.29 11.61 10.56 5 RNN, 7 total 1510 8.59 7.61 10.77 9.78 7 RNN, 9 total 1280 8.76 7.68 10.83 9.52 Table 1: Comparison of WER on a training and development set for various depths of RNN, with and without BatchNorm. The number of parameters is kept constant as the depth increases, thus the number of hidden units per layer decreases. All networks have 38 million parameters. The architecture “M RNN, N total” implies 1 layer of 1D convolution at the input, M consecutive bidirectional RNN layers, and the rest as fully-connected layers with N total layers in the network. 3.2 Batch Normalization for Deep RNNs To efficiently scale our model as we scale the training set, we increase the depth of the networks by adding more hidden layers, rather than making each layer larger. Previous work has examined doing so by increasing the number of consecutive bidirectional recurrent layers [24]. We explore Batch

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

19

adding more hidden layers, rather than making each layer larger. Previous work has examined doing so by increasing the number of consecutive bidirectional recurrent layers [24]. We explore Batch Normalization (BatchNorm) as a technique to accelerate training for such networks [63] since they often suffer from optimization issues. Recent research has shown that BatchNorm improves the speed of convergence of recurrent nets, without showing any improvement in generalization performance [37]. In contrast, we demonstrate that when applied to very deep networks of simple RNNs on large data sets, batch normalization substantially improves final generalization error while greatly accelerating training. In a typical feed-forward layer containing an affine transformation followed by a non-linearity f(
), we insert a BatchNorm transformation by applying f(B(Wh))instead off(Wh+b), where B(x) = xE[x] (Var[x] +)1=2+. (6) The terms EandVar are the empirical mean and variance over a minibatch. The bias bof the layer is dropped since its effect is cancelled by mean removal. The learnable parameters and allow the layer to scale and shift each hidden unit as desired. The constant is small and positive,

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

20

allow the layer to scale and shift each hidden unit as desired. The constant is small and positive, and is included only for numerical stability. In our convolutional layers the mean and variance are estimated over all the temporal output units for a given convolutional filter on a minibatch. The BatchNorm transformation reduces internal covariate shift by insulating a given layer from potentially uninteresting changes in the mean and variance of the layer’s input. We consider two methods of extending BatchNorm to bidirectional RNNs [37]. A natural extension is to insert a BatchNorm transformation immediately before every non-linearity. Equation 3 then becomes !hl t=f(B(Wlhl1 t+ !Ul !hl t1)). (7) In this case the mean and variance statistics are accumulated over a single time-step of the minibatch. The sequential dependence between time-steps prevents averaging over all time-steps. We find that this technique does not lead to improvements in optimization. We also tried accumulating an average over successive time-steps, so later time-steps are normalized over all present and previous timesteps. This also proved ineffective and greatly complicated backpropagation.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

21

over successive time-steps, so later time-steps are normalized over all present and previous timesteps. This also proved ineffective and greatly complicated backpropagation. We find that sequence-wise normalization [37] overcomes these issues. The recurrent computation is given by !hl t=f(B(Wlhl1 t) + !Ul !hl t1). (8) For each hidden unit, we compute the mean and variance statistics over all items in the minibatch over the length of the sequence. Figure 2 shows that deep networks converge faster with sequencewise normalization. Table 1 shows that the performance improvement from sequence-wise normalization increases with the depth of the network, with a 12% performance difference for the deepest network. When comparing depth, in order to control for model size we hold constant the total 6 50 100 150 200 250 300 Iteration ( ⇥103)2030405060Cost5-1 BN 5-1 No BN 9-7 BN 9-7 No BNFigure 2: Training curves of two models trained with and without BatchNorm. We start the plot after the first epoch of training as the curve is more difficult to interpret due to the SortaGrad curriculum method mentioned in Section 3.3 Train Dev

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

22

epoch of training as the curve is more difficult to interpret due to the SortaGrad curriculum method mentioned in Section 3.3 Train Dev Baseline BatchNorm Baseline BatchNorm Not Sorted 10.71 8.04 11.96 9.78 Sorted 8.76 7.68 10.83 9.52 Table 2: Comparison of WER on a training and development set with and without SortaGrad, and with and without batch normalization. number of parameters and still see strong performance gains. We would expect to see even larger improvements from depth if we held constant the number of activations per layer and added layers. We also find that BatchNorm harms generalization error for the shallowest network just as it converges slower for shallower networks. The BatchNorm approach works well in training, but is difficult to implement for a deployed ASR system, since it is often necessary to evaluate a single utterance in deployment rather than a batch. We find that normalizing each neuron to its mean and variance over just the sequence degrades performance. Instead, we store a running average of the mean and variance for the neuron collected during training, and use these for evaluation in deployment [63]. Using this technique, we can

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

23

performance. Instead, we store a running average of the mean and variance for the neuron collected during training, and use these for evaluation in deployment [63]. Using this technique, we can evaluate a single utterance at a time with better results than evaluating with a large batch. 3.3 SortaGrad Training on examples of varying length pose some algorithmic challenges. One possible solution is truncating backpropagation through time [68], so that all examples have the same sequence length during training [52]. However, this can inhibit the ability to learn longer term dependencies. Other works have found that presenting examples in order of difficulty can accelerate online learning [6, 70]. A common theme in many sequence learning problems including machine translation and speech recognition is that longer examples tend to be more challenging [11]. The CTC cost function that we use implicitly depends on the length of the utterance, L(x,y;) =logX `2Align (x,y)TY tpctc(`tjx;). (9) where Align (x,y)is the set of all possible alignments of the characters of the transcription yto frames of input xunder the CTC operator. In equation 9, the inner term is a product over time-steps

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

24

frames of input xunder the CTC operator. In equation 9, the inner term is a product over time-steps of the sequence, which shrinks with the length of the sequence since pctc(`tjx;)<1. This motivates a curriculum learning strategy we title SortaGrad. SortaGrad uses the length of the utterance as a heuristic for difficulty, since long utterances have higher cost than short utterances. 7 Architecture Simple RNN GRU 5 layers, 1 Recurrent 14.40 10.53 5 layers, 3 Recurrent 10.56 8.00 7 layers, 5 Recurrent 9.78 7.79 9 layers, 7 Recurrent 9.52 8.19 Table 3: Comparison of development set WER for networks with either simple RNN or GRU, for various depths. All models have batch normalization, one layer of 1D-invariant convolution, and approximately 38 million parameters. In the first training epoch, we iterate through the training set in increasing order of the length of the longest utterance in the minibatch. After the first epoch, training reverts back to a random order over minibatches. Table 2 shows a comparison of training cost with and without SortaGrad on the

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

25

over minibatches. Table 2 shows a comparison of training cost with and without SortaGrad on the 9 layer model with 7 recurrent layers. This effect is particularly pronounced for networks without BatchNorm, since they are numerically less stable. In some sense the two techniques substitute for one another, though we still find gains when applying SortaGrad and BatchNorm together. Even with BatchNorm we find that this curriculum improves numerical stability and sensitivity to small changes in training. Numerical instability can arise from different transcendental function implementations in the CPU and the GPU, especially when computing the CTC cost. This curriculum gives comparable results for both implementations. We suspect that these benefits occur primarily because long utterances tend to have larger gradients, yet we use a fixed learning rate independent of utterance length. Furthermore, longer utterances are more likely to cause the internal state of the RNNs to explode at an early stage in training. 3.4 Comparison of simple RNNs and GRUs The models we have shown so far are simple RNNs that have bidirectional recurrent layers with the recurrence for both the forward in time and backward in time directions modeled by Equation 3.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

26

recurrence for both the forward in time and backward in time directions modeled by Equation 3. Current research in speech and language processing has shown that having a more complex recurrence can allow the network to remember state over more time-steps while making them more computationally expensive to train [52, 8, 62, 2]. Two commonly used recurrent architectures are the Long Short-Term Memory (LSTM) units [30] and the Gated Recurrent Units (GRU) [11], though many other variations exist. A recent comprehensive study of thousands of variations of LSTM and GRU architectures showed that a GRU is comparable to an LSTM with a properly initialized forget gate bias, and their best variants are competitive with each other [32]. We decided to examine GRUs because experiments on smaller data sets showed the GRU and LSTM reach similar accuracy for the same number of parameters, but the GRUs were faster to train and less likely to diverge. The GRUs we use are computed by zt=(Wzxt+Uzht1+bz) rt=(Wrxt+Urht1+br) ~ht=f(Whxt+rtUhht1+bh) ht= (1zt)ht1+zt~ht(10)

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

27

~ht=f(Whxt+rtUhht1+bh) ht= (1zt)ht1+zt~ht(10) where(
)is the sigmoid function, zandrrepresent the update andreset gates respectively, and we drop the layer superscripts for simplicity. We differ slightly from the standard GRU in that we multiply the hidden state ht1byUhprior to scaling by the reset gate. This allows for all operations onht1to be computed in a single matrix multiplication. The output nonlinearity f(
)is typically the hyperbolic tangent function tanh . However, we find similar performance for tanh and clippedReLU nonlinearities and choose to use the clipped-ReLU for simplicity and uniformity with the rest of the network. Both GRU and simple RNN architectures benefit from batch normalization and show strong results with deep networks. However, Table 3 shows that for a fixed number of parameters, the GRU architectures achieve better WER for all network depths. This is clear evidence of the long term dependencies inherent in the speech recognition task present both within individual words and be8 Architecture Channels Filter dimension Stride Regular Dev Noisy Dev

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

28

dependencies inherent in the speech recognition task present both within individual words and be8 Architecture Channels Filter dimension Stride Regular Dev Noisy Dev 1-layer 1D 1280 11 2 9.52 19.36 2-layer 1D 640, 640 5, 5 1, 2 9.67 19.21 3-layer 1D 512, 512, 512 5, 5, 5 1, 1, 2 9.20 20.22 1-layer 2D 32 41x11 2x2 8.94 16.22 2-layer 2D 32, 32 41x11, 21x11 2x2, 2x1 9.06 15.71 3-layer 2D 32, 32, 96 41x11, 21x11, 21x11 2x2, 2x1, 2x1 8.61 14.74 Table 4: Comparison of WER for various arrangements of convolutional layers. In all cases, the convolutions are followed by 7 recurrent layers and 1 fully connected layer. For 2D-invariant convolutions the first dimension is frequency and the second dimension is time. All models have BatchNorm, SortaGrad, and 35 million parameters. tween words. As we discuss in Section 3.8, even simple RNNs are able to implicitly learn a language

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

29

parameters. tween words. As we discuss in Section 3.8, even simple RNNs are able to implicitly learn a language model due to the large amount of training data. Interestingly, the GRU networks with 5 or more recurrent layers do not significantly improve performance. We attribute this to the thinning from 1728 hidden units per layer for 1 recurrent layer to 768 hidden units per layer for 7 recurrent layers, to keep the total number of parameters constant. The GRU networks outperform the simple RNNs in Table 3. However, in later results (Section 6) we find that as we scale up the model size, for a fixed computational budget the simple RNN networks perform slightly better. Given this, most of the remaining experiments use the simple RNN layers rather than the GRUs. 3.5 Frequency Convolutions Temporal convolution is commonly used in speech recognition to efficiently model temporal translation invariance for variable length utterances. This type of convolution was first proposed for neural networks in speech more than 25 years ago [67]. Many neural network speech models have a first layer that processes input frames with some context window [16, 66]. This can be viewed as a temporal convolution with a stride of one.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

30

first layer that processes input frames with some context window [16, 66]. This can be viewed as a temporal convolution with a stride of one. Additionally, sub-sampling is essential to make recurrent neural networks computationally tractable with high sample-rate audio. The DS1 system accomplished this through the use of a spectrogram as input and temporal convolution in the first layer with a stride parameter to reduce the number of time-steps [26]. Convolutions in frequency and time domains, when applied to the spectral input features prior to any other processing, can slightly improve ASR performance [1, 53, 60]. Convolution in frequency attempts to model spectral variance due to speaker variability more concisely than what is possible with large fully connected networks. Since spectral ordering of features is removed by fullyconnected and recurrent layers, frequency convolutions work better as the first layers of the network. We experiment with adding between one and three layers of convolution. These are both in the timeand-frequency domain (2D invariance) and in the time-only domain (1D invariance). In all cases we use a same convolution, preserving the number of input features in both frequency and time. In some cases, we specify a stride across either dimension which reduces the size of the output. We do

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

31

use a same convolution, preserving the number of input features in both frequency and time. In some cases, we specify a stride across either dimension which reduces the size of the output. We do not explicitly control for the number of parameters, since convolutional layers add a small fraction of parameters to our networks. All networks shown in Table 4 have about 35 million parameters. We report results on two datasets—a development set of 2048 utterances (“Regular Dev”) and a much noisier dataset of 2048 utterances (“Noisy Dev”) randomly sampled from the CHiME 2015 development datasets [4]. We find that multiple layers of 1D-invariant convolutions provides a very small benefit. The 2D-invariant convolutions improve results substantially on noisy data, while providing a small benefit on clean data. The change from one layer of 1D-invariant convolution to three layers of 2D-invariant convolution improves WER by 23.9% on the noisy development set. 9 Dev no LM Dev LM Stride Unigrams Bigrams Unigrams Bigrams 2 14.93 14.56 9.52 9.66 3 15.01 15.60 9.65 10.06 4 18.86 14.84 11.92 9.93

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

32

3 15.01 15.60 9.65 10.06 4 18.86 14.84 11.92 9.93 Table 5: Comparison of WER with different amounts of striding for unigram and bigram outputs on a model with 1 layer of 1D-invariant convolution, 7 recurrent layers, and 1 fully connected layer. All models have BatchNorm, SortaGrad, and 35 million parameters. The models are compared on a development set with and without the use of a 5-gram language model. 3.6 Striding In the convolutional layers, we apply a longer stride and wider context to speed up training as fewer time-steps are required to model a given utterance. Downsampling the input sound (through FFT and convolutional striding) reduces the number of time-steps and computation required in the following layers, but at the expense of reduced performance. In our Mandarin models, we employ striding in the straightforward way. However, in English, striding can reduce accuracy simply because the output of our network requires at least one timestep per output character, and the number of characters in English speech per time-step is high enough to cause problems when striding2. To overcome this, we can enrich the English alphabet

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

33

enough to cause problems when striding2. To overcome this, we can enrich the English alphabet with symbols representing alternate labellings like whole words, syllables or non-overlapping ngrams. In practice, we use non-overlapping bi-graphemes or bigrams, since these are simple to construct, unlike syllables, and there are few of them compared to alternatives such as whole words. We transform unigram labels into bigram labels through a simple isomorphism. Non-overlapping bigrams shorten the length of the output transcription and thus allow for a decrease in the length of the unrolled RNN. The sentence the cat sat with non-overlapping bigrams is segmented as [th,e,space ,ca,t,space ,sa,t]. Notice that for words with odd number of characters, the last character becomes an unigram and space is treated as an unigram as well. This isomorphism ensures that the same words are always composed of the same bigram and unigram tokens. The output set of bigrams consists of all bigrams that occur in the training set. In Table 5 we show results for both the bigram and unigram systems for various levels of striding, with or without a language model. We observe that bigrams allow for larger strides without any

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

34

with or without a language model. We observe that bigrams allow for larger strides without any sacrifice in in the word error rate. This allows us to reduce the number of time-steps of the unrolled RNN benefiting both computation and memory usage. 3.7 Row Convolution and Unidirectional Models Bidirectional RNN models are challenging to deploy in an online, low-latency setting, because they are built to operate on an entire sample, and so it is not possible to perform the transcription process as the utterance streams from the user. We have found an unidirectional architecture that performs as well as our bidirectional models. This allows us to use unidirectional, forward-only RNN layers in our deployment system. To accomplish this, we employ a special layer that we call row convolution, shown in Figure 3. The intuition behind this layer is that we only need a small portion of future information to make an accurate prediction at the current time-step. Suppose at time-step t, we use a future context of  steps. We now have a feature matrix ht:t+= [ht,ht+1, ...,ht+]of sized(+ 1) . We define a

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

35

parameter matrix Wof the same size as ht:t+. The activations rtfor the new layer at time-step t are 2Chinese characters are more similar to English syllables than English characters. This is reflected in our training data, where there are on average 14.1 characters/s in English, while only 3.3 characters/s in Mandarin. Conversely, the Shannon entropy per character as calculated from occurrence in the training set, is less in English due to the smaller character set—4.9 bits/char compared to 12.6 bits/char in Mandarin. This implies that spoken Mandarin has a lower temporal entropy density, 41 bits/s compared to 58 bits/s, and can thus more easily be temporally compressed without losing character information. 10 Recurrent layerRow conv layerhtht+1ht+2ht+3rt+3rt+2rt+1rtFigure 3: Row convolution architecture with future context size of 2 rt,i=+1X j=1Wi,jht+j1,i, for 1id. (11) Since the convolution-like operation in Eq. 11 is row oriented for both Wandht:t+, we call this layer row convolution.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

36

Since the convolution-like operation in Eq. 11 is row oriented for both Wandht:t+, we call this layer row convolution. We place the row convolution layer above all recurrent layers. This has two advantages. First, this allows us to stream all computation below the row convolution layer on a finer granularity given little future context is needed. Second, this results in better CER compared to the best bidirectional model for Mandarin. We conjecture that the recurrent layers have learned good feature representations, so the row convolution layer simply gathers the appropriate information to feed to the classifier. Results for a unidirectional Mandarin speech system with row convolution and a comparison to a bidirectional model are given in Section 7 on deployment. 3.8 Language Model We train our RNN Models over millions of unique utterances, which enables the network to learn a powerful implicit language model. Our best models are quite adept at spelling, without any external language constraints. Further, in our development datasets we find many cases where our models can implicitly disambiguate homophones—for example, “he expects the Japanese agent tosell it for two hundred seventy five thousand dollars”. Nevertheless, the labeled training data is small compared

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

37

hundred seventy five thousand dollars”. Nevertheless, the labeled training data is small compared to the size of unlabeled text corpora that are available. Thus we find that WER improves when we supplement our system with a language model trained from external text. We use an n-gram language model since they scale well to large amounts of unlabeled text [26]. For English, our language model is a Kneser-Ney smoothed 5-gram model with pruning that is trained using the KenLM toolkit [28] on cleaned text from the Common Crawl Repository3. The vocabulary is the most frequently used 400,000 words from 250 million lines of text, which produces a language model with about 850 million n-grams. For Mandarin, the language model is a KneserNey smoothed character level 5-gram model with pruning that is trained on an internal text corpus of 8 billion lines of text. This produces a language model with about 2 billion n-grams. During inference we search for the transcription ythat maximizes Q(y)shown in Equation 12. This is a linear combination of logprobabilities from the CTC trained network and language model, along with a word insertion term [26].

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

38

is a linear combination of logprobabilities from the CTC trained network and language model, along with a word insertion term [26]. Q(y) = log(pctc(yjx)) +log(plm(y)) +word_count (y) (12) The weightcontrols the relative contributions of the language model and the CTC network. The weightencourages more words in the transcription. These parameters are tuned on a development set. We use a beam search to find the optimal transcription [27]. 3http://commoncrawl.org 11 Language Architecture Dev no LM Dev LM English 5-layer, 1 RNN 27.79 14.39 English 9-layer, 7 RNN 14.93 9.52 Mandarin 5-layer, 1 RNN 9.80 7.13 Mandarin 9-layer, 7 RNN 7.55 5.81 Table 6: Comparison of WER for English and CER for Mandarin with and without a language model. These are simple RNN models with only one layer of 1D invariant convolution. Table 6 shows that an external language model helps both English and Mandarin speech systems. The relative improvement given by the language model drops from 48% to 36% in English and 27%

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

39

Table 6 shows that an external language model helps both English and Mandarin speech systems. The relative improvement given by the language model drops from 48% to 36% in English and 27% to 23% in Mandarin, as we go from a model with 5 layers and 1 recurrent layer to a model with 9 layers and 7 recurrent layers. We hypothesize that the network builds a stronger implicit language model with more recurrent layers. The relative performance improvement from a language model is higher in English than in Mandarin. We attribute this to the fact that a Chinese character represents a larger block of information than an English character. For example, if we output directly to syllables or words in English, the model would make fewer spelling mistakes and the language model would likely help less. 3.9 Adaptation to Mandarin The techniques that we have described so far can be used to build an end-to-end Mandarin speech recognition system that outputs Chinese characters directly. This precludes the need to construct a pronunciation model, which is often a fairly involved component for porting speech systems to other languages [59]. Direct output to characters also precludes the need to explicitly model language specific pronunciation features. For example we do not need to model Mandarin tones explicitly, as some speech systems must do [59, 45]. The only architectural changes we make to our networks are due to the characteristics of the Chinese

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

40

some speech systems must do [59, 45]. The only architectural changes we make to our networks are due to the characteristics of the Chinese character set. Firstly, the output layer of the network outputs about 6000 characters, which includes the Roman alphabet, since hybrid Chinese-English transcripts are common. We incur an out of vocabulary error at evaluation time if a character is not contained in this set. This is not a major concern, as our test set has only 0.74% out of vocab characters. We use a character level language model in Mandarin as words are not usually segmented in text. The word insertion term of Equation 12 becomes a character insertion term. In addition, we find that the performance of the beam search during decoding levels off at a smaller beam size. This allows us to use a beam size of 200 with a negligible degradation in CER. In Section 6.2, we show that our Mandarin speech models show roughly the same improvements to architectural changes as our English speech models. 4 System Optimizations Our networks have tens of millions of parameters, and the training algorithm takes tens of singleprecision exaFLOPs to converge. Since our ability to evaluate hypotheses about our data and models depends on the ability to train models quickly, we built a highly optimized training system.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

41

This system has two main components—a deep learning library written in C ++, along with a highperformance linear algebra library written in both CUDA and C ++. Our optimized software, running on dense compute nodes with 8 Titan X GPUs per node, allows us to sustain 24 single-precision teraFLOP/second when training a single model on one node. This is 45% of the theoretical peak computational throughput of each node. We also can scale to multiple nodes, as outlined in the next subsection. 4.1 Scalability and Data-Parallelism We use the standard technique of data-parallelism to train on multiple GPUs using synchronous SGD. Our most common configuration uses a minibatch of 512on8GPUs. Our training pipeline 12 binds one process to each GPU. These processes then exchange gradient matrices during the backpropagation by using all-reduce, which exchanges a matrix between multiple processes and sums the result so that at the end, each process has a copy of the sum of all matrices from all processes. We find synchronous SGD useful because it is reproducible and deterministic. We have found that the appearance of non-determinism in our system often signals a serious bug, and so having reproducibility as a goal has greatly facilitates debugging. In contrast, asynchronous methods such

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

42

that the appearance of non-determinism in our system often signals a serious bug, and so having reproducibility as a goal has greatly facilitates debugging. In contrast, asynchronous methods such as asynchronous SGD with parameter servers as found in Dean et al. [18] typically do not provide reproducibility and are therefore more difficult to debug. Synchronous SGD is simple to understand and implement. It scales well as we add multiple nodes to the training process. 2021222324252627 GPUs211212213214215216217218219Time (seconds)5-3 (2560) 9-7 (1760) Figure 4: Scaling comparison of two networks—a 5 layer model with 3 recurrent layers containing 2560 hidden units in each layer and a 9 layer model with 7 recurrent layers containing 1760 hidden units in each layer. The times shown are to train 1 epoch. The 5 layer model trains faster because it uses larger matrices and is more computationally efficient. Figure 4 shows that time taken to train one epoch halves as we double the number of GPUs that we train on, thus achieving near-linear weak scaling. We keep the minibatch per GPU constant at 64 during this experiment, effectively doubling the minibatch as we double the number of GPUs.

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

43

64 during this experiment, effectively doubling the minibatch as we double the number of GPUs. Although we have the ability to scale to large minibatches, we typically use either 8 or 16 GPUs during training with a minibatch of 512 or 1024, in order to converge to the best result. Since all-reduce is critical to the scalability of our training, we wrote our own implementation of the ring algorithm [48, 65] for higher performance and better stability. Our implementation avoids extraneous copies between CPU and GPU, and is fundamental to our scalability. We configure OpenMPI with the smcuda transport that can send and receive buffers residing in the memory of two different GPUs by using GPUDirect. When two GPUs are in the same PCI root complex, this avoids any unnecessary copies to CPU memory. This also takes advantage of tree-structured interconnects by running multiple segments of the ring concurrently between neighboring devices. We built our implementation using MPI send and receive, along with CUDA kernels for the elementwise operations. Table 7 compares the performance of our all-reduce implementation with that provided by OpenMPI version 1.8.5. We report the time spent in all-reduce for a full training run that ran for one epoch on our English dataset using a 5 layer, 3 recurrent layer architecture with 2560 hidden units for all

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

44

on our English dataset using a 5 layer, 3 recurrent layer architecture with 2560 hidden units for all layers. In this table, we use a minibatch of 64 per GPU, expanding the algorithmic minibatch as we scale to more GPUs. We see that our implementation is considerably faster than OpenMPI’s when the communication is within a node (8 GPUs or less). As we increase the number of GPUs and increase the amount of inter-node communication, the gap shrinks, although our implementation is still 2-4X faster. All of our training runs use either 8 or 16 GPUs, and in this regime, our all-reduce implementation results in 2.5faster training for the full training run, compared to using OpenMPI directly. Optimizing all-reduce has thus resulted in important productivity benefits for our experiments, and has made our simple synchronous SGD approach scalable. 13 GPU OpenMPI Our Performance all-reduce all-reduce Gain 4 55359.1 2587.4 21.4 8 48881.6 2470.9 19.8 16 21562.6 1393.7 15.5 32 8191.8 1339.6 6.1 64 1395.2 611.0 2.3 128 1602.1 422.6 3.8

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

45

32 8191.8 1339.6 6.1 64 1395.2 611.0 2.3 128 1602.1 422.6 3.8 Table 7: Comparison of two different all-reduce implementations. All times are in seconds. Performance gain is the ratio of OpenMPI all-reduce time to our all-reduce time. Language Architecture CPU CTC Time GPU CTC Time Speedup English 5-layer, 3 RNN 5888.12 203.56 28.9 Mandarin 5-layer, 3 RNN 1688.01 135.05 12.5 Table 8: Comparison of time spent in seconds in computing the CTC loss function and gradient in one epoch for two different implementations. Speedup is the ratio of CPU CTC time to GPU CTC time. 4.2 GPU implementation of CTC loss function Calculating the CTC loss function is more complicated than performing forward and back propagation on our RNN architectures. Originally, we transferred activations from the GPUs to the CPU, where we calculated the loss function using an OpenMP parallelized implementation of CTC. However, this implementation limited our scalability rather significantly, for two reasons. Firstly, it became computationally more significant as we improved efficiency and scalability of the RNN

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

46

it became computationally more significant as we improved efficiency and scalability of the RNN itself. Secondly, transferring large activation matrices between CPU and GPU required us to spend interconnect bandwidth for CTC, rather than on transferring gradient matrices to allow us to scale using data parallelism to more processors. To overcome this, we wrote a GPU implementation of the CTC loss function. Our parallel implementation relies on a slight refactoring to simplify the dependences in the CTC calculation, as well as the use of optimized parallel sort implementations from ModernGPU [5]. We give more details of this parallelization in the Appendix. Table 8 compares the performance of two CTC implementations. The GPU implementation saves us 95 minutes per epoch in English, and 25 minutes in Mandarin. This reduces overall training time by 10-20%, which is also an important productivity benefit for our experiments. 4.3 Memory allocation Our system makes frequent use of dynamic memory allocations to GPU and CPU memory, mainly to store activation data for variable length utterances, and for intermediate results. Individual allocations can be very large; over 1 GB for the longest utterances. For these very large allocations we found that CUDA’s memory allocator and even std::malloc introduced significant overhead

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

47

we found that CUDA’s memory allocator and even std::malloc introduced significant overhead into our application—over a 2x slowdown from using std::malloc in some cases. This is because both cudaMalloc andstd::malloc forward very large allocations to the operating system or GPU driver to update the system page tables. This is a good optimization for systems running multiple applications, all sharing memory resources, but editing page tables is pure overhead for our system where nodes are dedicated entirely to running a single model. To get around this limitation, we wrote our own memory allocator for both CPU and GPU allocations. Our implementation follows the approach of the last level shared allocator in jemalloc: all allocations are carved out of contiguous memory blocks using the buddy algorithm [34]. To avoid fragmentation, we preallocate all of GPU memory at the start of training and subdivide individual allocations from this block. Similarly, we set the CPU memory block size that we forward to mmap to be substantially larger than std::malloc , at 12GB. 14 Dataset Speech Type Hours WSJ read 80 Switchboard conversational 300 Fisher conversational 2000 LibriSpeech read 960 Baidu read 5000 Baidu mixed 3600 Total 11940

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

48

WSJ read 80 Switchboard conversational 300 Fisher conversational 2000 LibriSpeech read 960 Baidu read 5000 Baidu mixed 3600 Total 11940 Table 9: Summary of the datasets used to train DS2 in English. The Wall Street Journal (WSJ), Switchboard and Fisher [13] corpora are all published by the Linguistic Data Consortium. The LibriSpeech dataset [46] is available free on-line. The other datasets are internal Baidu corpora. Most of the memory required for training deep recurrent networks is used to store activations through each layer for use by back propagation, not to store the parameters of the network. For example, storing the weights for a 70M parameter network with 9 layers requires approximately 280 MB of memory, but storing the activations for a batch of 64, seven-second utterances requires 1.5 GB of memory. TitanX GPUs include 12GB of GDDR5 RAM, and sometimes very deep networks can exceed the GPU memory capacity when processing long utterances. This can happen unpredictably, especially when the distribution of utterance lengths includes outliers, and it is desirable to avoid a catastrophic failure when this occurs. When a requested memory allocation exceeds available GPU

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208

1512.02595

49

especially when the distribution of utterance lengths includes outliers, and it is desirable to avoid a catastrophic failure when this occurs. When a requested memory allocation exceeds available GPU memory, we allocate page-locked GPU-memory-mapped CPU memory using cudaMallocHost instead. This memory can be accessed directly by the GPU by forwarding individual memory transactions over PCIe at reduced bandwidth, and it allows a model to continue to make progress even after encountering an outlier. The combination of fast memory allocation with a fallback mechanism that allows us to slightly overflow available GPU memory in exceptional cases makes the system significantly simpler, more robust, and more efficient. 5 Training Data Large-scale deep learning systems require an abundance of labeled training data. We have collected an extensive training dataset for both English and Mandarin speech models, in addition to augmenting our training with publicly available datasets. In English we use 11,940 hours of labeled speech data containing 8 million utterances summarized in Table 9. For the Mandarin system we use 9,400 hours of labeled audio containing 11 million utterances. The Mandarin speech data consists of internal Baidu corpora, representing a mix of read speech and spontaneous speech, in both standard Mandarin and accented Mandarin. 5.1 Dataset Construction

1512.02595

Deep Speech 2: End-to-End Speech Recognition in English and Mandarin

We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.

http://arxiv.org/pdf/1512.02595

cs.CL

20151208

20151208