Video-OPD: Efficient Post-Training of Multimodal Large Language Models for Temporal Video Grounding via On-Policy Distillation

Benchmark Model Rank Results
temporal-sentence-grounding-on-activitynet-timelensVideo-OPD 8B (Round 3) (extra data)–mIoU: 50.7R@1 IoU=0.3: 64.4R@1 IoU=0.5: 49.2R@1 IoU=0.7: 39.0
temporal-sentence-grounding-on-qvhighlights-timelens-validationVideo-OPD 8B (Round 3) (extra data)–mIoU: 64.6R@1 IoU=0.3: 77.6R@1 IoU=0.5: 65.3R@1 IoU=0.7: 55.3