ResearchBenchmarksVisual Evidence Quality Evaluation on M3CoT Reasoning (subset of 500 samples)Follow76.4AIM-CoT Win RateAIM-CoT72.5874.4976.478.31Sep 30, 2025Evaluation ResultsMethodMethodLinksAIM-CoT Win RateTie RateICoT Win RateAIM-CoTSelection Strategy=Inf...Selection Strategy=Info. Gain-driven AVP, Judge=GPT-4v2025.0976.416.27.4ICoTSelection Strategy=Att...Selection Strategy=Attention-driven Top-K, Judge=GPT-4v2025.0976.416.27.4