When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language Models
MOTIVE uses multi-view self-verification so vision-language models accept reliable answers and rethink uncertain ones.
Vision-language models often emit plausible but incorrect answers, and single-criterion self-verification is unstable across prompts. MOTIVE evaluates candidates from complementary views and learns a correctness-aligned reliability score that decides whether to accept an answer or trigger history-guided rethinking. Across multimodal benchmarks and VLM backbones it outperforms self-verification and self-correction baselines while reducing unnecessary reasoning turns, without an external judge.
- MOTIVE scores answers from complementary verification views.
- A learned reliability score decides whether to accept or rethink.
- No external judge is required during inference.
- The method beats self-verification baselines and cuts extra reasoning turns.
From the original article
1 min read at huggingface.co
Vision-language models (VLMs) have achieved strong performance in multimodal reasoning, yet they remain prone to generating plausible but incorrect answers. Self-verification offers a practical way to improve answer reliability without relying on external judges, but existing methods typically depend on a single verification criterion or fixed prompt, resulting in incomplete and unstable reliability estimates.
By Ziquan Zhu, Hanruo Zhu, Si-Yuan Lu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.