Reproduce “Learning to Correct Geometry in Generated Videos” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weijie Lyu, Tiancheng Shen, Jason Li, Yujing Wang, Xiaoyuan Wang, Haobo Yuan, Yizhou Zhao, Ming-Hsuan Yang
VGC is a separate video-to-video corrector that repairs geometric distortions in generated clips without changing the source generator. It is trained on paired examples made by distorting camera-motion videos, with adaptive keyframes helping clips follow target camera paths. On DL3DV, CogVideoX 1.5-5B results improve in PSNR, SSIM, MSE, and 3DCS, but worsen in LPIPS; evaluation compares reconstructed frames, not ground-truth video.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.