1 of 1

1. Ablation study of LPIPS loss and stereo consistency loss
2. Ablation study of different strategies

SC-NAFSSR: Perceptual-Oriented Stereo Image Super-Resolution Using

Stereo Consistency Guided NAFSSR�Zidian Qiu, Zongyao He, Zhihao Zhan, Zilin Pan, Xingyuan Xian, and Zhi Jin*

School of Intelligent Systems Engineering, Sun Yat-sen University

​

Framework

Ablation Study

Results

Contributions

Methods

Code: https://github.com/FVL2020/SC-NAFSSR

Contact: Zhi Jin jinzh26@mail.sysu.edu.cn

Ablation study of LPIPS loss and stereo consistency loss

Ablation study of different strategies

Qualitative results

Quantitative results

Applying GAN = less regular artifacts

• Based on the NTIRE 2022 Stereo Image Super-Resolution Challenge champion scheme NAFSSR, we combine the LPIPS perceptual loss with the traditional VGG perceptual loss for training.
• To improve stereo consistency, we perform supervision on each SCAM.
• Our method without GAN end up ranking second place on the Flickr1024 test set.

Note: The blue arrows indicate supervision during the training process.