README.md
5.6 KB · 104 lines · markdown Raw
1 ---
2 license: apache-2.0
3 tags:
4 - vision
5 - image-classification
6 datasets:
7 - imagenet-1k
8 - imagenet-21k
9 widget:
10 - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/tiger.jpg
11 example_title: Tiger
12 - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/teapot.jpg
13 example_title: Teapot
14 - src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/palace.jpg
15 example_title: Palace
16 ---
17
18 # Vision Transformer (base-sized model)
19
20 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.
21
22 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.
23
24 ## Model description
25
26 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, also at resolution 224x224.
27
28 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.
29
30 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.
31
32 ## Intended uses & limitations
33
34 You can use the raw model for image classification. See the [model hub](https://huggingface.co/models?search=google/vit) to look for
35 fine-tuned versions on a task that interests you.
36
37 ### How to use
38
39 Here is how to use this model to classify an image of the COCO 2017 dataset into one of the 1,000 ImageNet classes:
40
41 ```python
42 from transformers import ViTImageProcessor, ViTForImageClassification
43 from PIL import Image
44 import requests
45
46 url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
47 image = Image.open(requests.get(url, stream=True).raw)
48
49 processor = ViTImageProcessor.from_pretrained('google/vit-base-patch16-224')
50 model = ViTForImageClassification.from_pretrained('google/vit-base-patch16-224')
51
52 inputs = processor(images=image, return_tensors="pt")
53 outputs = model(**inputs)
54 logits = outputs.logits
55 # model predicts one of the 1000 ImageNet classes
56 predicted_class_idx = logits.argmax(-1).item()
57 print("Predicted class:", model.config.id2label[predicted_class_idx])
58 ```
59
60 For more code examples, we refer to the [documentation](https://huggingface.co/transformers/model_doc/vit.html#).
61
62 ## Training data
63
64 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.
65
66 ## Training procedure
67
68 ### Preprocessing
69
70 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).
71
72 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).
73
74 ### Pretraining
75
76 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. Training resolution is 224.
77
78 ## Evaluation results
79
80 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.
81
82 ### BibTeX entry and citation info
83
84 ```bibtex
85 @misc{wu2020visual,
86 title={Visual Transformers: Token-based Image Representation and Processing for Computer Vision},
87 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},
88 year={2020},
89 eprint={2006.03677},
90 archivePrefix={arXiv},
91 primaryClass={cs.CV}
92 }
93 ```
94
95 ```bibtex
96 @inproceedings{deng2009imagenet,
97 title={Imagenet: A large-scale hierarchical image database},
98 author={Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li},
99 booktitle={2009 IEEE conference on computer vision and pattern recognition},
100 pages={248--255},
101 year={2009},
102 organization={Ieee}
103 }
104 ```