ViT

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

PaddlePaddle training/validation code and pretrained models for ViT.

The official TF implementation is here.

This implementation is developed by PaddleViT.

ViT Model Overview

Update

• Update (2022-03-15): Code is refactored, old weights link are updated, more weights are uploaded.
• Update (2021-09-27): More weights are uploaded.
• Update (2021-08-11): Code is released and ported weights are uploaded.

Models Zoo

Model	Acc@1	Acc@5	#Params	FLOPs	Image Size	Crop_pct	Interpolation	Link
vit_tiny_patch16_224	75.48	92.84	5.7M	1.1G	224	0.875	bicubic	google/baidu
vit_tiny_patch16_384	78.42	94.54	5.7M	3.3G	384	1.0	bicubic	google/baidu
vit_small_patch32_224	76.23	93.35	22.9M	1.1G	224	0.875	bicubic	google/baidu
vit_small_patch32_384	80.48	95.60	22.9M	3.3G	384	1.0	bicubic	google/baidu
vit_small_patch16_224	81.40	96.15	22.0M	4.3G	224	0.875	bicubic	google/baidu
vit_small_patch16_384	83.80	97.10	22.0M	12.7G	384	1.0	bicubic	google/baidu
vit_base_patch32_224	80.68	95.61	88.2M	4.4G	224	0.875	bicubic	google/baidu
vit_base_patch32_384	83.35	96.84	88.2M	12.7G	384	1.0	bicubic	google/baidu
vit_base_patch16_224	84.58	97.30	86.4M	17.0G	224	0.875	bicubic	google/baidu
vit_base_patch16_384	85.99	98.00	86.4M	49.8G	384	1.0	bicubic	google/baidu
vit_large_patch32_384	81.51	96.09	306.5M	44.4G	384	1.0	bicubic	google/baidu
vit_large_patch16_224	85.81	97.82	304.1M	59.9G	224	0.875	bicubic	google/baidu
vit_large_patch16_384	87.08	98.30	304.1M	175.9G	384	1.0	bicubic	google/baidu

*The results are evaluated on ImageNet2012 validation set.

Note: old model weights may not be corrected loaded to current version, please download and use the current model weights.

Data Preparation

ImageNet2012 dataset is used in the following file structure:

│imagenet/
├──train_list.txt
├──val_list.txt
├──train/
│  ├── n01440764
│  │   ├── n01440764_10026.JPEG
│  │   ├── n01440764_10027.JPEG
│  │   ├── ......
│  ├── ......
├──val/
│  ├── n01440764
│  │   ├── ILSVRC2012_val_00000293.JPEG
│  │   ├── ILSVRC2012_val_00002138.JPEG
│  │   ├── ......
│  ├── ......

• train_list.txt: list of relative paths and labels of training images. You can download it from: google/baidu
• val_list.txt: list of relative paths and labels of validation images. You can download it from: google/baidu

Usage

To use the model with pretrained weights, download the .pdparam weight file and change related file paths in the following python scripts. The model config files are located in ./configs/.

For example, assume weight file is downloaded in ./vit_base_patch16_224.pdparams, to use the vit_base_patch16_224 model in python:

from config import get_config
from vit import build_vit as build_model
# config files in ./configs/
config = get_config('./configs/vit_base_patch16_224.yaml')
# build model
model = build_model(config)
# load pretrained weights
model_state_dict = paddle.load('./vit_base_patch16_224.pdparams')
model.set_state_dict(model_state_dict)

Evaluation

To evaluate ViT model performance on ImageNet2012, run the following script using command line:

sh run_eval_multi.sh

or

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
python main_multi_gpu.py \
-cfg='./configs/vit_tiny_patch16_224.yaml' \
-dataset='imagenet2012' \
-batch_size=256 \
-data_path='/dataset/imagenet' \
-eval \
-pretrained='./vit_tiny_patch16_224.pdparams' \
-amp

Note: if you have only 1 GPU, change device number to CUDA_VISIBLE_DEVICES=0 would run the evaluation on single GPU.

Training

To train the ViT model on ImageNet2012, run the following script using command line:

sh run_train_multi.sh

or

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
python main_multi_gpu.py \
-cfg='./configs/vit_base_patch16_224.yaml' \
-dataset='imagenet2012' \
-batch_size=128 \
-data_path='/dataset/imagenet' \
-amp

Note: it is highly recommanded to run the training using multiple GPUs / multi-node GPUs.

Finetuning

To finetune the ViT model on ImageNet2012, run the following script using command line:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
python main_multi_gpu.py \
-cfg='./configs/vit_base_patch16_224.yaml' \
-dataset='imagenet2012' \
-batch_size=128 \
-data_path='/dataset/imagenet' \
-pretrained='./vit_base_patch16_224.pdparams' \
-amp

Note: use -pretrained argument to set the pretrained model path, you may also need to modify the hyperparams defined in config file.

Reference

@article{dosovitskiy2020image,
  title={An image is worth 16x16 words: Transformers for image recognition at scale},
  author={Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and others},
  journal={arXiv preprint arXiv:2010.11929},
  year={2020}
}