.. Copyright 2021 The HuggingFace Team. All rights reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. DETR ----------------------------------------------------------------------------------------------------------------------- Overview ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The DETR model was proposed in `End-to-End Object Detection with Transformers `__ by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov and Sergey Zagoruyko. DETR consists of a convolutional backbone followed by an encoder-decoder Transformer which can be trained end-to-end for object detection. It greatly simplifies a lot of the complexity of models like Faster-R-CNN and Mask-R-CNN, which use things like region proposals, non-maximum suppression procedure and anchor generation. Moreover, DETR can also be naturally extended to perform panoptic segmentation, by simply adding a mask head on top of the decoder outputs. The abstract from the paper is the following: *We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries, DETR reasons about the relations of the objects and the global image context to directly output the final set of predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and highly-optimized Faster RCNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive baselines.* This model was contributed by `nielsr `__. The original code can be found `here `__. The quickest way to get started with DETR is by checking the `example notebooks `__ (which showcase both inference and fine-tuning on custom data). Here's a TLDR explaining how :class:`~transformers.DetrForObjectDetection` works: First, an image is sent through a pre-trained convolutional backbone (in the paper, the authors use ResNet-50/ResNet-101). Let's assume we also add a batch dimension. This means that the input to the backbone is a tensor of shape :obj:`(batch_size, 3, height, width)`, assuming the image has 3 color channels (RGB). The CNN backbone outputs a new lower-resolution feature map, typically of shape :obj:`(batch_size, 2048, height/32, width/32)`. This is then projected to match the hidden dimension of the Transformer of DETR, which is :obj:`256` by default, using a :obj:`nn.Conv2D` layer. So now, we have a tensor of shape :obj:`(batch_size, 256, height/32, width/32).` Next, the feature map is flattened and transposed to obtain a tensor of shape :obj:`(batch_size, seq_len, d_model)` = :obj:`(batch_size, width/32*height/32, 256)`. So a difference with NLP models is that the sequence length is actually longer than usual, but with a smaller :obj:`d_model` (which in NLP is typically 768 or higher). Next, this is sent through the encoder, outputting :obj:`encoder_hidden_states` of the same shape (you can consider these as image features). Next, so-called **object queries** are sent through the decoder. This is a tensor of shape :obj:`(batch_size, num_queries, d_model)`, with :obj:`num_queries` typically set to 100 and initialized with zeros. These input embeddings are learnt positional encodings that the authors refer to as object queries, and similarly to the encoder, they are added to the input of each attention layer. Each object query will look for a particular object in the image. The decoder updates these embeddings through multiple self-attention and encoder-decoder attention layers to output :obj:`decoder_hidden_states` of the same shape: :obj:`(batch_size, num_queries, d_model)`. Next, two heads are added on top for object detection: a linear layer for classifying each object query into one of the objects or "no object", and a MLP to predict bounding boxes for each query. The model is trained using a **bipartite matching loss**: so what we actually do is compare the predicted classes + bounding boxes of each of the N = 100 object queries to the ground truth annotations, padded up to the same length N (so if an image only contains 4 objects, 96 annotations will just have a "no object" as class and "no bounding box" as bounding box). The `Hungarian matching algorithm `__ is used to find an optimal one-to-one mapping of each of the N queries to each of the N annotations. Next, standard cross-entropy (for the classes) and a linear combination of the L1 and `generalized IoU loss `__ (for the bounding boxes) are used to optimize the parameters of the model. DETR can be naturally extended to perform panoptic segmentation (which unifies semantic segmentation and instance segmentation). :class:`~transformers.DetrForSegmentation` adds a segmentation mask head on top of :class:`~transformers.DetrForObjectDetection`. The mask head can be trained either jointly, or in a two steps process, where one first trains a :class:`~transformers.DetrForObjectDetection` model to detect bounding boxes around both "things" (instances) and "stuff" (background things like trees, roads, sky), then freeze all the weights and train only the mask head for 25 epochs. Experimentally, these two approaches give similar results. Note that predicting boxes is required for the training to be possible, since the Hungarian matching is computed using distances between boxes. Tips: - DETR uses so-called **object queries** to detect objects in an image. The number of queries determines the maximum number of objects that can be detected in a single image, and is set to 100 by default (see parameter :obj:`num_queries` of :class:`~transformers.DetrConfig`). Note that it's good to have some slack (in COCO, the authors used 100, while the maximum number of objects in a COCO image is ~70). - The decoder of DETR updates the query embeddings in parallel. This is different from language models like GPT-2, which use autoregressive decoding instead of parallel. Hence, no causal attention mask is used. - DETR adds position embeddings to the hidden states at each self-attention and cross-attention layer before projecting to queries and keys. For the position embeddings of the image, one can choose between fixed sinusoidal or learned absolute position embeddings. By default, the parameter :obj:`position_embedding_type` of :class:`~transformers.DetrConfig` is set to :obj:`"sine"`. - During training, the authors of DETR did find it helpful to use auxiliary losses in the decoder, especially to help the model output the correct number of objects of each class. If you set the parameter :obj:`auxiliary_loss` of :class:`~transformers.DetrConfig` to :obj:`True`, then prediction feedforward neural networks and Hungarian losses are added after each decoder layer (with the FFNs sharing parameters). - If you want to train the model in a distributed environment across multiple nodes, then one should update the `num_boxes` variable in the `DetrLoss` class of `modeling_detr.py`. When training on multiple nodes, this should be set to the average number of target boxes across all nodes, as can be seen in the original implementation `here `__. - :class:`~transformers.DetrForObjectDetection` and :class:`~transformers.DetrForSegmentation` can be initialized with any convolutional backbone available in the `timm library `__. Initializing with a MobileNet backbone for example can be done by setting the :obj:`backbone` attribute of :class:`~transformers.DetrConfig` to :obj:`"tf_mobilenetv3_small_075"`, and then initializing the model with that config. - DETR resizes the input images such that the shortest side is at least a certain amount of pixels while the longest is at most 1333 pixels. At training time, scale augmentation is used such that the shortest side is randomly set to at least 480 and at most 800 pixels. At inference time, the shortest side is set to 800. One can use :class:`~transformers.DetrFeatureExtractor` to prepare images (and optional annotations in COCO format) for the model. Due to this resizing, images in a batch can have different sizes. DETR solves this by padding images up to the largest size in a batch, and by creating a pixel mask that indicates which pixels are real/which are padding. Alternatively, one can also define a custom :obj:`collate_fn` in order to batch images together, using :meth:`~transformers.DetrFeatureExtractor.pad_and_create_pixel_mask`. - The size of the images will determine the amount of memory being used, and will thus determine the :obj:`batch_size`. It is advised to use a batch size of 2 per GPU. See `this Github thread `__ for more info. As a summary, consider the following table: +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Task** | **Object detection** | **Instance segmentation** | **Panoptic segmentation** | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Description** | Predicting bounding boxes and class labels around | Predicting masks around objects (i.e. instances) in an image | Predicting masks around both objects (i.e. instances) as well as | | | objects in an image | | "stuff" (i.e. background things like trees and roads) in an image | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Model** | :class:`~transformers.DetrForObjectDetection` | :class:`~transformers.DetrForSegmentation` | :class:`~transformers.DetrForSegmentation` | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Example dataset** | COCO detection | COCO detection, | COCO panoptic | | | | COCO panoptic | | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Format of annotations to provide to** | {‘image_id’: int, | {‘image_id’: int, | {‘file_name: str, | | :class:`~transformers.DetrFeatureExtractor` | ‘annotations’: List[Dict]}, each Dict being a COCO | ‘annotations’: [List[Dict]] } (in case of COCO detection) | ‘image_id: int, | | | object annotation | | ‘segments_info’: List[Dict] } | | | | or | | | | | | and masks_path (path to directory containing PNG files of the masks) | | | | {‘file_name’: str, | | | | | ‘image_id’: int, | | | | | ‘segments_info’: List[Dict]} (in case of COCO panoptic) | | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **Postprocessing** (i.e. converting the | :meth:`~transformers.DetrFeatureExtractor.post_process` | :meth:`~transformers.DetrFeatureExtractor.post_process_segmentation` | :meth:`~transformers.DetrFeatureExtractor.post_process_segmentation`, | | output of the model to COCO API) | | | :meth:`~transformers.DetrFeatureExtractor.post_process_panoptic` | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ | **evaluators** | :obj:`CocoEvaluator` with iou_types = “bbox” | :obj:`CocoEvaluator` with iou_types = “bbox”, “segm” | :obj:`CocoEvaluator` with iou_tupes = “bbox, “segm” | | | | | | | | | | :obj:`PanopticEvaluator` | +---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+ In short, one should prepare the data either in COCO detection or COCO panoptic format, then use :class:`~transformers.DetrFeatureExtractor` to create :obj:`pixel_values`, :obj:`pixel_mask` and optional :obj:`labels`, which can then be used to train (or fine-tune) a model. For evaluation, one should first convert the outputs of the model using one of the postprocessing methods of :class:`~transformers.DetrFeatureExtractor`. These can be be provided to either :obj:`CocoEvaluator` or :obj:`PanopticEvaluator`, which allow you to calculate metrics like mean Average Precision (mAP) and Panoptic Quality (PQ). The latter objects are implemented in the `original repository `__. See the `example notebooks `__ for more info regarding evaluation. DETR specific outputs ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.models.detr.modeling_detr.DetrModelOutput :members: .. autoclass:: transformers.models.detr.modeling_detr.DetrObjectDetectionOutput :members: .. autoclass:: transformers.models.detr.modeling_detr.DetrSegmentationOutput :members: DetrConfig ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.DetrConfig :members: DetrFeatureExtractor ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.DetrFeatureExtractor :members: __call__, pad_and_create_pixel_mask, post_process, post_process_segmentation, post_process_panoptic DetrModel ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.DetrModel :members: forward DetrForObjectDetection ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.DetrForObjectDetection :members: forward DetrForSegmentation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. autoclass:: transformers.DetrForSegmentation :members: forward