Reproduce “Guided Interpretable Facial Expression Recognition via Spatial Action Unit Cues” 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
Soufiane Belharbi, Marco Pedersoli, Alessandro Lameiras Koerich, Simon Bacon, Eric Granger
The authors train facial-expression classifiers with an expression-specific action-unit guide derived from facial landmarks and labels, adding an alignment loss that encourages internal attention maps to match estimated facial-region heatmaps. On RAF-DB, APViT accuracy changes from 91.00% to 91.03%, while layer-5 ATT-COS rises from 0.38 to 0.85. This measures proxy-map agreement in this comparison, not validated interpretability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.