V-Rubrics: Rule-Based RL Boosts VLM Visual Faithfulness

NTU's V-Rubrics decomposes a single score into fine-grained criteria like visual faithfulness and uses rule-based reinforcement learning to fix VLMs' credit-assignment failures.

2026-08-27 ~ 2026-08-27 · 2 related posts

• V-Rubrics: Visual Faithfulness via Rubric-Based RL — nanyang-technological-university-singapore · 2026-08-27
• V-Rubrics: Improving Visual Faithfulness via Rubric-Based Reinforcement Learning — liuziwei7 · 2026-08-27