EventLens: Event-Structure Reinforcement Learning for Video Understanding

Benchmark Model Rank Results
video-question-answering-on-tempcompass-overallEventLens-8B#1Accuracy: 77.17
video-question-answering-on-videoholmesEventLens-8B#4Accuracy: 59.44
temporal-sentence-grounding-on-qvhighlights-timelens-validationEventLens-8B#12mIoU: 59.12
temporal-sentence-grounding-on-activitynet-timelensEventLens-8B#13mIoU: 46.35
video-question-answering-on-video-mmmuEventLens-8B#32Accuracy: 39.6
video-question-answering-on-mvbenchEventLens-8B#51Avg.: 65.42
moment-retrieval-on-charades-staEventLens-8B#80mIoU: 48.25