Reproduce “Benchmarking the Explanatory Quality of Open-Weight Vision-Language Models in Face Recognition” 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
Laurent Colbois, Sébastien Marcel
The authors evaluate open-weight vision-language models for forensic face comparison with structured JSON explanations listing feature comparisons alongside similarity scores. They score relevance and occlusion-focused faithfulness using automated lexical checks, separately from verification performance. On structured-output LFW, Gemma3-27B reports 5.2% EER versus ArcFace’s 0.3%; similar matching can mask explanation-quality differences, and these benchmark results do not establish operational forensic performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.