Reproduce “Supervised Descent Method for Solving Nonlinear Least Squares Problems in Computer Vision” 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

Xuehan Xiong, Fernando De la Torre

SDM learns sequences of parameter updates to optimize nonlinear least-squares objectives without computing Jacobians or Hessians at test time. It trains maps from feature mismatches to corrections. For face alignment, it updates landmark locations using local HOG features. On LFPW, its mean error was 3.47 × 10⁻², comparable to Cao et al.’s 3.43 × 10⁻², though unavailable images limited fairness.

Summary by alphaXiv.

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