mmpretrain/configs/vision_transformer

Vision Transformer

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Introduction

Vision Transformer, known as ViT, succeeded in using a full transformer to outperform previous works that based on convolutional networks in vision field. ViT splits image into patches to feed the multi-head attentions, concatenates a learnable class token for final prediction and adds a learnable position embeddings for relative positional message between patches. Based on these three techniques with attentions, ViT provides a brand-new pattern to build a basic structure in vision field.

The strategy works even better when coupled with large datasets pre-trainings. Because of its simplicity and effectiveness, some after works in classification field are originated from ViT. And even in recent multi-modality field, ViT-based method still plays a role in it.

Abstract

Show the paper's abstract

While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train.

How to use it?

Predict image

>>> import torch
>>> from mmcls.apis import init_model, inference_model
>>>
>>> model = init_model('configs/vision_transformer/vit-base-p16_pt-32xb128-mae_in1k-224.py', 'https://download.openmmlab.com/mmclassification/v0/vit/vit-base-p16_pt-32xb128-mae_in1k_20220623-4c544545.pth')
>>> predict = inference_model(model, 'demo/demo.JPEG')
>>> print(predict['pred_class'])
sea snake
>>> print(predict['pred_score'])
0.9184340238571167

Use the model

>>> import torch
>>> from mmcls.apis import init_model
>>>
>>> model = init_model('configs/vision_transformer/vit-base-p16_pt-32xb128-mae_in1k-224.py', 'https://download.openmmlab.com/mmclassification/v0/vit/vit-base-p16_pt-32xb128-mae_in1k_20220623-4c544545.pth')
>>> inputs = torch.rand(1, 3, 224, 224).to(model.data_preprocessor.device)
>>> # To get classification scores.
>>> out = model(inputs)
>>> print(out.shape)
torch.Size([1, 1000])
>>> # To extract features.
>>> outs = model.extract_feat(inputs)
>>> # The patch token features
>>> print(outs[0][0].shape)
torch.Size([1, 768, 14, 14])
>>> # The cls token features
>>> print(outs[0][1].shape)
torch.Size([1, 768])

Train/Test Command

Place the ImageNet dataset to the data/imagenet/ directory, or prepare datasets according to the docs.

Train:

python tools/train.py configs/vision_transformer/vit-base-p16_pt-32xb128-mae_in1k-224.py

Test:

python tools/test.py configs/vision_transformer/vit-base-p16_pt-32xb128-mae_in1k-224.py https://download.openmmlab.com/mmclassification/v0/vit/vit-base-p16_pt-32xb128-mae_in1k_20220623-4c544545.pth

For more configurable parameters, please refer to the API.

Results and models

The training step of Vision Transformers is divided into two steps. The first step is training the model on a large dataset, like ImageNet-21k, and get the pre-trained model. And the second step is training the model on the target dataset, like ImageNet-1k, and get the fine-tuned model. Here, we provide both pre-trained models and fine-tuned models.

ImageNet-21k

The pre-trained models on ImageNet-21k are used to fine-tune, and therefore don't have evaluation results.

Model	resolution	Params(M)	Flops(G)	Download
ViT-B16*	224x224	86.86	33.03	model
ViT-B32*	224x224	88.30	8.56	model
ViT-L16*	224x224	304.72	116.68	model

Models with * are converted from the official repo.

ImageNet-1k

Model	Pretrain	resolution	Params(M)	Flops(G)	Top-1 (%)	Top-5 (%)	Config	Download
ViT-B16	From scratch	224x224	86.86	33.03	82.37	96.15	config	model | log
ViT-B16*	ImageNet-21k	384x384	86.86	33.03	85.43	97.77	config	model
ViT-B16 (IPU)	ImageNet-21k	224x224	86.86	33.03	81.22	95.56	config	model | log
ViT-B32*	ImageNet-21k	384x384	88.30	8.56	84.01	97.08	config	model
ViT-L16*	ImageNet-21k	384x384	304.72	116.68	85.63	97.63	config	model

Models with * are converted from the official repo. The config files of these models are only for validation. We don't ensure these config files' training accuracy and welcome you to contribute your reproduction results.

Citation

@inproceedings{
  dosovitskiy2021an,
  title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author={Alexey Dosovitskiy and Lucas Beyer and Alexander Kolesnikov and Dirk Weissenborn and Xiaohua Zhai and Thomas Unterthiner and Mostafa Dehghani and Matthias Minderer and Georg Heigold and Sylvain Gelly and Jakob Uszkoreit and Neil Houlsby},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=YicbFdNTTy}
}