NTIRE 2022 Challenge on Efficient Super-Resolution: Methods and Results

Li, Yawei; Zhang, Kai; Timofte, Radu; Gool, Luc Van; Kong, Fangyuan; Li, Mingxi; Liu, Songwei; Du, Zongcai; Liu, Ding; Zhou, Changcong; Chen, Jingyi; Han, Qingrui; Li, Zheyuan; Liu, Yingqi; Chen, Xiangyu; Cai, Haoming; Qiao, Yu; Dong, Chao; Sun, Lei; Pan, Jinshan; Zhu, Yi; Zong, Zhikai; Hui, Zheng; Yang, Tao; Ren, Peiran; Xie, Xuansong; Hua, Xian-Sheng; Li, Han; Ji, Xiaozhong; Lin, Chuming; Luo, Donghao; Tai, Ying; Wang, Chengjie; Zhang, Zhizhong; Xie, Yuan; Shen, Cheng,; Luo, Ziwei; Yu, Lei; Wen, Zhihong; Wu, Qi; Li, Youwei; Fan, Haoqiang; Sun, Jian; Liu, Shuaicheng; Huang, Yuanfei; Jin, Meiguang; Huang, Hua; Liu, Jing; Zhang, Xinjian; Wang, Yan; Long, Lingshun; Li, Gen; Zhang, Tracy; Cao, Zhigang; Sun, Lei; Panaetov, Alexander; Wang, Yucong; Cai, Minjie; Wang, Li; Lan, Tian; Wang, Zheyuan; Ma, Hongbing; Liu, Jie; Chen, Chao; Cai, Yao; Tang, Jie; Wu, Guorong; Wang, Weiran; Huang, Shirui; Lu, Honglei; Liu, Huan; Wang, Keyan; Chen, Jun; Chen, Shi; Miao, Youzhi; Huang, Zeyu; Zhang, Lefei; Ayazoğlu, Mustafa; Xiong, Wei; Xiong, Caiming; Wang, Fei; Li, Hao; Wen, Ruimian; Yang, Zhijing; Zou, Wenbin; Zheng, Weixin; Tian, Ye; Zhang, Yuncheng; Kong, Xiangzhen; Arora, Aditya; Zamir, Syed Waqas; Khan, Salman; Hayat, Munawar; Khan, Fahad Shahbaz; Ning, Dandan Gaoand Dengwen Zhouand Qian; Jingzhu, Tang,; Huang, Han; Wang, Yufei; Peng, Zhangheng; Li, Haobo; Guan, Wenxue; Gong, Shenghua; Li, Xin; Li, Jun; Wang, Wanjun; Zhou, Dengwen; Zeng, Kun; Lin, Hanjiang; Chen, Xinyu; Fang, Jinsheng

doi:10.1109/cvprw56347.2022.00118

Cited by 40 publications

(6 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, our VapSR ×4 uses 21.68 % and 28.18 % parameters of RFDN ×4 and IMDN ×4, while obtains average 0.187 dB improvement on four evaluation datasets. Moreover, the proposed VapSR-S achieves competitive performance to BSRN-S [35], which is the winner of the model complexity sub-track in NTIRE 2022 Challenge on Efficient Super-Resolution [34]. Interestingly, our structure has relative advantages on metric SSIM than PSNR as well.…”

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 90%

“…RFDN [38] rethinks the channel splitting operation and introduces the progressive refinement module as an equivalent architecture. In NTIRE 2022 Efficient SR Challenge [34], RLFN [28] won the championship in the runtime track by ditching the multi-branch design of RFDN and introducing a contrastive loss for faster computation and better performance. BSRN [35] won the first place in the model complexity track by replacing the standard convolution with a well-designed depth-wise separable convolution to save computations and utilizing two effective attention schemes to enhance the model ability.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Efficient Image Super-Resolution using Vast-Receptive-Field Attention

Zhang¹,

Cai²,

Gu³

et al. 2022

Preprint

View full text Add to dashboard Cite

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 depth-wise 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. The code and models are available at https://github.com/zhoumumu/VapSR.

show abstract

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 90%

Section: Related Workmentioning

confidence: 99%

Efficient Image Super-Resolution using Vast-Receptive-Field Attention

Zhang¹,

Cai²,

Gu³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…RFDN [9] rethought the information distillation structure and proposed a feature distillation connection operation (FDC) that was equivalent to channel separation operation, which was more lightweight and flexible, and introduced a progressive refinement module as an equivalent architecture. In the NTIRE 2022 efficient SR challenge [16], RLFN [17] abandoned the multi-branch design of RFDN and introduced contrast loss, which resulted in faster computation speed and better performance, and won the championship in the runtime track.…”

Section: Lightweight Super-resolution Networkmentioning

confidence: 99%

A lightweight image super-resolution network based on large receptive field information distillation

wu,

wang,

shen

et al. 2024

Second International Conference on Informatics, Networking, and Computing (ICINC 2023)

View full text Add to dashboard Cite

This paper tackles the problem of network structure redundancy in image super-resolution (SR) algorithms and aims to develop an efficient and low-parameter SR method. It proposes a lightweight SR network based on vast-receptive-field information distillation (VIDN), which enhances the model performance and reduces the parameter size by introducing large receptive field convolution and attention modules, optimizing convolution operations, and using simple gate functions instead of RELU activation functions. The network leverages large kernel convolution and multi-scale attention mechanisms to better capture global information and fuse local and global features of the image, respectively. The ESA and CCA modules complement each other to improve image clarity and realism, forming a powerful multi-scale attention mechanism. The experimental results on three scales and four benchmark datasets demonstrate that the proposed algorithm achieves an average PSNR improvement of 0.03db, an average parameter reduction of 53%, and clearer and more natural visual effects on images.

show abstract

“…This challenge is one of the NTIRE 2022 associated challenges: spectral recovery [7], spectral demosaicing [6], perceptual image quality assessment [19], inpainting [40], night photography rendering [17], efficient super-resolution [31], learning the super-resolution space [34], super-resolution and quality enhancement of compressed video [49], high dynamic range [38], stereo super-resolution [45], burst super-resolution [8].…”

Section: Introductionmentioning

confidence: 99%

NTIRE 2022 Spectral Recovery Challenge and Data Set

Arad¹,

Timofte²,

Yahel³

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self Cite

View full text Add to dashboard Cite

This paper reviews the third biennial challenge on spectral reconstruction from RGB images, i.e., the recovery of whole-scene hyperspectral (HS) information from a 3-channel RGB image. This challenge presents the "ARAD 1K" data set: a new, larger-than-ever natural hyperspectral image data set containing 1,000 images. Challenge participants were required to recover hyperspectral information from synthetically generated JPEGcompressed RGB images simulating capture by a known calibrated camera, operating under partially known parameters, in a setting which includes acquisition noise. The challenge was attended by 241 teams, with 60 teams competing in the final testing phase, 12 of which provided detailed descriptions of their methodology which are included in this report. The performance of these submissions is reviewed and provided here as a gauge for the current stateof-the-art in spectral reconstruction from natural RGB images.

show abstract

NTIRE 2022 Challenge on Efficient Super-Resolution: Methods and Results

Cited by 40 publications

References 53 publications

Efficient Image Super-Resolution using Vast-Receptive-Field Attention

Efficient Image Super-Resolution using Vast-Receptive-Field Attention

A lightweight image super-resolution network based on large receptive field information distillation

NTIRE 2022 Spectral Recovery Challenge and Data Set

Contact Info

Product

Resources

About