nielsr HF staff commited on
Commit
a485839
1 Parent(s): 7b9a687

First draft of model card

Browse files
Files changed (1) hide show
  1. README.md +95 -0
README.md ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ datasets:
4
+ - cifar10
5
+ - cifar100
6
+ - imagenet
7
+ - imagenet-21k
8
+ - oxford-iiit-pets
9
+ - oxford-flowers-102
10
+ - vtab
11
+ ---
12
+
13
+ # Vision Transformer base model
14
+
15
+ Vision Transformer (ViT) model pre-trained on ImageNet-21k (14 million images, 21,843 classes) at resolution 224x224, and fine-tuned on ImageNet 2012 (1 million images, 1,000 classes) at resolution 224x224. It was introduced in the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Dosovitskiy et al. and first released in [this repository](https://github.com/google-research/vision_transformer). However, the weights were converted from the [timm repository](https://github.com/rwightman/pytorch-image-models) by Ross Wightman, who already converted the weights from JAX to PyTorch. Credits go to him.
16
+
17
+ Disclaimer: The team releasing ViT did not write a model card for this model so this model card has been written by the Hugging Face team.
18
+
19
+ ## Model description
20
+
21
+ The Vision Transformer (ViT) is a transformer encoder model (BERT-like) pretrained on a large collection of images in a supervised fashion, namely ImageNet-21k, at a resolution of 224x224 pixels. Next, the model was fine-tuned on ImageNet (also referred to as ILSVRC2012), a dataset comprising 1 million images and 1,000 classes, at the same resolution, 224x224.
22
+
23
+ Images are presented to the model as a sequence of fixed-size patches (resolution 16x16), which are linearly embedded. One also adds a [CLS] token to the beginning of a sequence to use it for classification tasks. One also adds absolute position embeddings before feeding the sequence to the layers of the Transformer encoder.
24
+
25
+ By pre-training the model, it learns an inner representation of images that can then be used to extract features useful for downstream tasks: if you have a dataset of labeled images for instance, you can train a standard classifier by placing a linear layer on top of the pre-trained encoder. One typically places a linear layer on top of the [CLS] token, as the last hidden state of this token can be seen as a representation of an entire image.
26
+
27
+ ## Intended uses & limitations
28
+
29
+ You can use the raw model for image classification. See the [model hub](https://huggingface.co/models?search=google/vit) to look for
30
+ fine-tuned versions on a task that interests you.
31
+
32
+ ### How to use
33
+
34
+ Here is how to use this model to classify an image of the COCO 2017 dataset into one of the 1,000 ImageNet classes:
35
+
36
+ ```python
37
+ from transformers import ViTFeatureExtractor, ViTForImageClassification
38
+ from PIL import Image
39
+ import requests
40
+ url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
41
+ image = Image.open(requests.get(url, stream=True).raw)
42
+ feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-large-patch16-224')
43
+ model = ViTForImageClassification.from_pretrained('google/vit-large-patch16-224')
44
+ inputs = feature_extractor(images=image)
45
+ outputs = model(**inputs)
46
+ logits = outputs.logits
47
+ # model predicts one of the 1000 ImageNet classes
48
+ predicted_class = logits.argmax(-1).item()
49
+ ```
50
+
51
+ Currently, both the feature extractor and model support PyTorch. Tensorflow and JAX/FLAX are coming soon, and the API of ViTFeatureExtractor might change.
52
+
53
+ ## Training data
54
+
55
+ The ViT model was pretrained on [ImageNet-21k](http://www.image-net.org/), a dataset consisting of 14 million images and 21k classes, and fine-tuned on [ImageNet](http://www.image-net.org/challenges/LSVRC/2012/), a dataset consisting of 1 million images and 1k classes.
56
+
57
+ ## Training procedure
58
+
59
+ ### Preprocessing
60
+
61
+ The exact details of preprocessing of images during training/validation can be found [here](https://github.com/google-research/vision_transformer/blob/master/vit_jax/input_pipeline.py).
62
+
63
+ Images are resized/rescaled to the same resolution (224x224) and normalized across the RGB channels with mean (0.5, 0.5, 0.5) and standard deviation (0.5, 0.5, 0.5).
64
+
65
+ ### Pretraining
66
+
67
+ The model was trained on TPUv3 hardware (8 cores). All model variants are trained with a batch size of 4096 and learning rate warmup of 10k steps. For ImageNet, the authors found it beneficial to additionally apply gradient clipping at global norm 1. Pre-training resolution is 224.
68
+
69
+ ## Evaluation results
70
+
71
+ For evaluation results on several image classification benchmarks, we refer to tables 2 and 5 of the original paper. Note that for fine-tuning, the best results are obtained with a higher resolution (384x384). Of course, increasing the model size will result in better performance.
72
+
73
+ ### BibTeX entry and citation info
74
+
75
+ ```bibtex
76
+ @misc{wu2020visual,
77
+ title={Visual Transformers: Token-based Image Representation and Processing for Computer Vision},
78
+ author={Bichen Wu and Chenfeng Xu and Xiaoliang Dai and Alvin Wan and Peizhao Zhang and Zhicheng Yan and Masayoshi Tomizuka and Joseph Gonzalez and Kurt Keutzer and Peter Vajda},
79
+ year={2020},
80
+ eprint={2006.03677},
81
+ archivePrefix={arXiv},
82
+ primaryClass={cs.CV}
83
+ }
84
+ ```
85
+
86
+ ```bibtex
87
+ @inproceedings{deng2009imagenet,
88
+ title={Imagenet: A large-scale hierarchical image database},
89
+ author={Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li},
90
+ booktitle={2009 IEEE conference on computer vision and pattern recognition},
91
+ pages={248--255},
92
+ year={2009},
93
+ organization={Ieee}
94
+ }
95
+ ```