Reframe: Post-Training Framework Aligns Visual Agents' Observations

Reframe, a two-stage post-training framework from Nanjing University and CASIA, aligns visual agents' observation states rather than just output distributions, ensuring image evidence from tools is truly used. It surpasses SOTA on 14 benchmarks, beating Mix-RL and the original MOPD.

2026-11-10 ~ 2026-11-11 · 2 related posts

• Reframe: NJU's Two-Stage Post-Training Framework Aligns Visual Agents' Observations, Beats Mix-RL and Vanilla MOPD — NJU · 2026-11-10
• Reframe: Post-Training Framework Makes Visual Agents Actually Use What They See — _akhaliq · 2026-11-11