Haonan Ge, Yiwei Wang, Kai-Wei Chang, Hang Wu, Yujun Cai: FrameMind: Frame-Interleaved Chain-of-Thought for Video Reasoning via Reinforcement Learning https://arxiv.org/abs/2509.24008 https://arxiv.org/pdf/2509.24008 https://arxiv.org/html/2509.24008

FrameMind employs a reinforcement‑learning policy called DRFS‑GRPO to dynamically request video frames, and it achieves higher accuracy on the MLVU and VideoMME benchmarks. https://getnews.me/framemind-enables-adaptive-video-reasoning-with-reinforcement-learning/ #framemind #drfsgrpo #videoanalysis

Haonan Ge, Yiwei Wang, Kai-Wei Chang, Hang Wu, Yujun Cai
FrameMind: Frame-Interleaved Chain-of-Thought for Video Reasoning via Reinforcement Learning
https://arxiv.org/abs/2509.24008