arxivcs.CV2026-07-14
Gaussian Mixture Modeling for Event-Aware Visual Allocation in Long Video Understanding
Yifan Lu, Ziqi Zhang, Chunfeng Yuan, Jun Gao, Bing Li, Weiming Hu
Large Vision-Language Models (LVLMs) face significant challenges in long video understanding due to the excessive computational cost and information loss associated with uniform sampling. Existing keyframe selection methods often treat video frames as atomic entities and allocate…