Structured Causal Video Reasoning via Multi-Objective Alignment

Benchmark Model Rank Results
temporal-sentence-grounding-on-charades-timelensFactum-4B–R1@0.5: 40.4R1@0.7: 21.6
video-question-answering-on-tempcompass-overallSTEER-4B–Accuracy: 70.64
zero-shot-video-question-answer-on-video-mme-w-o-subsSTEER-4B–Accuracy (%): 64.7