CvT

CvT: Introducing Convolutions to Vision Transformers, arxiv

PaddlePaddle training/validation code and pretrained models for CvT.

The official pytorch implementation is here.

This implementation is developed by PaddleViT.

CvT Model Overview

Update

• Update (2022-04-08): Code is updated.
• Update (2021-12-24): Code is released and ported weights are uploaded.

Models Zoo

Model	Acc@1	Acc@5	#Params	FLOPs	Image Size	Crop_pct	Interpolation	Link
CvT-13-224	81.59	95.67	20M	4.5G	224	0.875	bicubic	google/baidu
CvT-21-224	82.46	96.00	32M	7.1G	224	0.875	bicubic	google/baidu
CvT-13-384	83.00	96.36	20M	16.3G	384	1.0	bicubic	google/baidu
CvT-21-384	83.27	96.16	32M	24.9G	384	1.0	bicubic	google/baidu
CvT-13-384-22k	83.26	97.09	20M	16.3G	384	1.0	bicubic	google/baidu
CvT-21-384-22k	84.91	97.62	32M	24.9G	384	1.0	bicubic	google/baidu
CvT-w24-384-22k	87.58	98.47	277M	193.2G	384	1.0	bicubic	google/baidu

*The results are evaluated on ImageNet2012 validation set.

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 ./cvt-13-224x224.pdparams, to use the cvt-13-224x224 model in python:

from config import get_config
from cvt import build_cvt as build_model
# config files in ./configs/
config = get_config('./configs/cvt-13-224x224.yaml')
# build model
model = build_model(config)
# load pretrained weights
model_state_dict = paddle.load('./cvt-13-224x224.pdparams')
model.set_state_dict(model_state_dict)

Evaluation

To evaluate 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/cvt-13-224x224.yaml' \
-dataset='imagenet2012' \
-batch_size=256 \
-data_path='/dataset/imagenet' \
-eval \
-pretrained='./cvt-13-224x224.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 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/cvt-13-224x224.yaml' \
-dataset='imagenet2012' \
-batch_size=256 \
-data_path='/dataset/imagenet' \
-amp

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

Reference

@article{wu2021cvt,
title={CvT: Introducing Convolutions to Vision Transformers},
author={Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang},
journal={arXiv preprint arXiv:2103.15808},
year={2021}
}