Reproduce “Deep multi-scale video prediction beyond mean square error” 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
Michael Mathieu, Camille Couprie, Yann LeCun
The authors predict future video frames using a coarse-to-fine model trained with pixel-level, adversarial, and gradient-difference losses. On moving regions of UCF101, the combined adversarial-plus-gradient-difference model scores higher than adversarial training alone on PSNR, SSIM, and sharpness-difference for the first future frame. Against optical flow, it scores higher on two measures but lower on SSIM.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.