Competition

Start: Feb. 6, 2025, 11 p.m.

Description: Project Page: https://github.com/lixinustc/KVQE-Challenge-CVPR-NTIRE2025. The submission example can be found in "https://drive.google.com/file/d/1xp5mHuZqkAFWUqEpKy2LlqoE2XTTp0do/view?usp=drive_link". The final score is computed with PSNR+10*SSIM-10*LPIPS+0.1*MUSIQ+10*ManIQA+10*CLIPIQA. The top three team will be selected from the top six teams on **final score** in the test stage based on user study (measuring perceptual fidelity and realism). If you have some problems on submission, please contact us through Forum. Please keep the same file name and architecture as example when submitting your results. The amount of the prize (??? dollars) will be determined soon. It is suggested to design diffusion-based methods since the provided LR image in the wild share the same resolution with HR while the LR in synthetic is 4x down-sampled from HR images. Please inference all images with the same model.