kapynResearch

LVSum: A Benchmark for Timestamp-Aware Long Video Summarization

LVSum is a new benchmark for evaluating timestamp-aware long video summarization in MLLMs. The human-annotated dataset includes 72 diverse videos across 13 domains, averaging 16 minutes in length with fine-grained temporal references. Developers working on multimodal video understanding can use this benchmark to test how well their models maintain temporal fidelity and semantic grounding over extended durations.

Apple ML Research·Jul 20, 2026

Opening Kapyn…