Efficient Progressive High Dynamic Range Image Restoration via Attention and Alignment Network

Gaocheng Yu, Jin Zhang, Zhe Ma, Hongbin Wang

Abstract: HDR is an important part of computational photography technology. In this paper, we propose a lightweight neural network called Efficient Attention-and-alignment-guided Progressive Network (EAPNet) for the challenge NTIRE 2022 HDR Track 1 and Track 2. We introduce a multi-scale lightweight encoding module to extract features. Besides, we propose Progressive Dilated U-shape Block (PDUB) which is a progressive plug-and-play module for dynamically tuning MAccs and PSNR. Finally, we use fast and low-power feature-alignment module to deal with misalignment problem in place of the time-consuming Deformable Convolutional Network (DCN). The experiments show that our method achieves about 20× compression on MAccs with better PSNR-µ and PSNR compared to the state-of-the-art method. We got the 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nd</sup> place of both two tracks during the testing phase. Fig. 1 shows the visualized result of NTIRE 2022 HDR challenge.