Reproduce “Efficient Video Face Enhancement with Enhanced Spatial-Temporal Consistency” 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

Yutong Wang, Jiajie Teng, Jiajiong Cao, Yuming Li, Chenguang Ma, Hongteng Xu, Dixin Luo

STC restores compressed video faces using separate codebooks for portrait appearance and temporal changes, predicting codes from low-quality clips. It combines temporal attention and motion residuals, with marginal prior regularization to improve codebook use. On VFHQ-Test, it reports SSIM 0.8641, FVD 105.1, and 2.995 seconds per 24-frame clip, versus PGTFormer’s 0.8426, 107.6, and 7.085 seconds; PGTFormer performs better on PSNR, LPIPS, and AKD.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.