MVP: Enhancing Video Large Language Models via Self-supervised Masked Video Prediction
This work addresses the problem of improving temporal understanding in Video Large Language Models for video analysis tasks, representing an incremental advancement in post-training methods.
The paper tackles the limitation of Video Large Language Models in capturing temporal coherence and inter-frame correlations by proposing a novel post-training objective called Masked Video Prediction (MVP), which enhances video reasoning capabilities by directly reinforcing temporal reasoning and causal understanding.
Reinforcement learning based post-training paradigms for Video Large Language Models (VideoLLMs) have achieved significant success by optimizing for visual-semantic tasks such as captioning or VideoQA. However, while these approaches effectively enhance perception abilities, they primarily target holistic content understanding, often lacking explicit supervision for intrinsic temporal coherence and inter-frame correlations. This tendency limits the models' ability to capture intricate dynamics and fine-grained visual causality. To explicitly bridge this gap, we propose a novel post-training objective: Masked Video Prediction (MVP). By requiring the model to reconstruct a masked continuous segment from a set of challenging distractors, MVP forces the model to attend to the sequential logic and temporal context of events. To support scalable training, we introduce a scalable data synthesis pipeline capable of transforming arbitrary video corpora into MVP training samples, and further employ Group Relative Policy Optimization (GRPO) with a fine-grained reward function to enhance the model's understanding of video context and temporal properties. Comprehensive evaluations demonstrate that MVP enhances video reasoning capabilities by directly reinforcing temporal reasoning and causal understanding.