Efficient Image Super-Resolution using Vast-Receptive-Field Attention(ECCVW 2022). Paper link

Lin Zhou1∗, Haoming Cai1∗, Jinjin Gu, Zheyuan Li, Yingqi Liu, Xiangyu Chen, Yu Qiao, Chao Dong (* indicates equal contribution)

The attention mechanism plays a pivotal role in designing advanced super-resolution (SR) networks. In this work, we design an efficient SR network by improving the attention mechanism. We start from a simple pixel attention module and gradually modify it to achieve better super-resolution performance with reduced parameters. The specific approaches include: (1) increasing the receptive field of the attention branch, (2) replacing large dense convolution kernels with depthwise separable convolutions, and (3) introducing pixel normalization. These approaches paint a clear evolutionary roadmap for the design of attention mechanisms. Based on these observations, we propose VapSR, the Vast-receptive-field Pixel attention network. Experiments demonstrate the superior performance of VapSR. VapSR outperforms the present lightweight networks with even fewer parameters. And the light version of VapSR can use only 21.68% and 28.18% parameters of IMDB and RFDN to achieve similar performances to those networks.

Quick Intro: the presentation and the poster.

Performance of X4 scale (PSNR / SSIM on Y channel):

The code is constructed based on BasicSR. Before any testing or reproducing, make sure the installation and the datasets preparation are done correctly.

To keep the workspace clean and simple, only test.py, train.py and your_arch.py are needed here and then you are good to go.

This line in test.py and train.py enables them to register your arch:

from vapsr import vapsr

You can run the testing demo with

>>> CUDA_VISIBLE_DEVICES=0 python code/test.py -opt options/test/VapSR_X4.yml

Reproduce with

>>> CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port=4321 code/train.py -opt options/train/VapSR_X4.yml --launcher pytorch

Feel free to contact us:

Lin Zhou📫(zhougrace885@gmail.com) 
Haoming Cai📫(hmcai@umd.edu) 
Jinjin Gu📫(jinjin.gu@sydney.edu.au) 
Zheyuan Li📫(zy.li3@siat.ac.cn) 
Yingqi Liu📫(yq.liu3@siat.ac.cn) 
Xiangyu Chen📫(chxy95@gmail.com) 
Yu Qiao📫(yu.qiao@siat.ac.cn) 
Chao Dong📫(chao.dong@siat.ac.cn)