On August 28, 2020, through the European Conference on Computer Vision (ECCV), the Advances in Image Manipulation Challenges (AIM2020) was held simultaneously with ECCV. Led by Prof. QingWang, a competing team composed of ZhongyunHu, Xin Huangand YaningLi from the Computer Vision and Computational Photography Laboratory of School of Computer Science (www.npu-cvpg.org), Northwestern Polytechnical University (NPU) won the Re-lighting Track championship of the competition.

Hosted by ETH Zurich, Switzerland and sponsored by international renowned high-tech giants such as Huawei, Qualcomm, Nvidia and Google, this competition had attracted more than 80 teams from home and abroad. The competition, which has 16 tracks this year, covers a number of hot research areas in image and video processing, including re-lighting, super-resolution, image completion, etc. The Re-lighting Track requires the participating groups to relight the target image using the illumination information in the source image to make the illumination information of the re-lightened image be aligned with that of the source image. Advances in re-lighting, which is widely used in augmented reality, visual effects and lighting compensation, will be a significant boost to the development of computer vision and computer graphics.

Under Prof. Wang's guidance, ZhongyunHu, a PhD student of our university, designed a relighting method that combines auto-encoder with the attention mechanism. With the help of the attention mechanism, the decoder will be dedicated to the rendering of areas related to relighting in the target image. ZhongyunHu introduced his method in the online Image Processing Progress Seminar due to the COVID-19 impact, which has attracted widespread attention among the organizers and other competing teams.

Written by: ZhongyunHu, Reviewed by: WuGao