Temporal Gains, Spatial Costs: Revisiting Video Fine-Tuning in Multimodal Large Language Models
This work addresses the challenge of balancing spatial and temporal understanding in multimodal large language models, which is crucial for improving visual capabilities in AI applications, though it is incremental in nature.
The study systematically investigates how video-based supervised fine-tuning (Video-SFT) affects multimodal large language models, finding that it reliably improves video performance but often leads to limited gains or degradation on static image benchmarks, with a trade-off tied to temporal budget. The authors propose an instruction-aware Hybrid-Frame strategy that partially mitigates this image-video trade-off.
Multimodal large language models (MLLMs) are typically trained in multiple stages, with video-based supervised fine-tuning (Video-SFT) serving as a key step for improving visual understanding. Yet its effect on the fine-grained evolution of visual capabilities, particularly the balance between spatial and temporal understanding, remains poorly understood. In this paper, we systematically study how Video-SFT reshapes visual capabilities in MLLMs. Across architectures, parameter scales, and frame sampling settings, we observe a consistent pattern: Video-SFT reliably improves video performance, but often yields limited gains or even degradation on static image benchmarks. We further show that this trade-off is closely tied to temporal budget: increasing the number of sampled frames generally improves video performance, but does not reliably improve static image performance. Motivated by this finding, we study an instruction-aware Hybrid-Frame strategy that adaptively allocates frame counts and partially mitigates the image-video trade-off. Our results indicate that Video-SFT is not a free lunch for MLLMs, and preserving spatial understanding remains a central challenge in joint image-video training.