New VIGOR benchmark tests visual rationale correctness in multimodal language models
A new arXiv paper introduces VIGOR, a benchmark that evaluates whether multimodal large language models provide correct visual rationales, not just correct answers.
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-11-11 05:00 · arXiv cs.CV
VIGOR: Benchmarking Visual Rationale Correctness in Multimodal Large Language Models