ReasonAct: Progressive Training for Fine-Grained Video Reasoning in Small Models
This work addresses the challenge of video reasoning for small models, offering incremental improvements in performance while maintaining computational efficiency.
The paper tackles the problem of fine-grained temporal reasoning in small-scale multimodal models for video understanding by introducing ReasonAct, a three-stage progressive training method, resulting in accuracy improvements of up to 17.9 points on benchmarks like HMDB51, UCF-101, and Kinetics-400.
While recent multimodal models have shown progress in vision-language tasks, small-scale variants still struggle with the fine-grained temporal reasoning required for video understanding. We introduce ReasonAct, a method that enhances video reasoning in smaller models through a three-stage training process: first building a foundation with text-only reasoning, then fine-tuning on video, and finally refining with temporal-aware reinforcement learning. We build upon Temporal Group Relative Policy Optimization (T-GRPO) by incorporating temporal consistency modeling into policy optimization. We also propose a biomechanically-motivated sub-action decomposition mechanism that provides graduated rewards for constituent action phases. Through experiments on HMDB51, UCF-101, and Kinetics-400, our 3B-parameter model achieves 67.2%, 94.1%, and 78.9% accuracy respectively, demonstrating improvements of 17.9, 15.8, and 12.3 points over baselines. Ablation studies validate that our progressive training methodology enables smaller models to achieve competitive video reasoning performance while maintaining computational efficiency.