This work proposes Counterfactual Relational Policy Optimization (CRPO), which explicitly trains Video LLMs to respond to controlled changes in the visual input, and introduces DyBench, a paired counterfactual benchmark with 3{,}014 videos and a strict pair-accuracy metric.
Da-Zhao Du, Jian Liu, Jia-Long Qin et al.· arXiv.org· 1 citation
WorldCycle is introduced, a self-verifiable RL framework that constructs closed action cycles and their repeated executions from ordinary action sequences, and optimizes two complementary rewards: a spatial closure reward enforcing symmetry between mirrored forward and reverse segments, and a temporal consistency rewar...
Bohai Gu, Yueyang Yuan, Tai-Yi Wu et al.· 2 citations
This work introduces CamChoreo, a benchmark of 4,229 real single-shot clips with expert-annotated temporal segments, and proposes CamDistill, which distills the same geometric knowledge into lightweight camera tokens during training and removes the 3D model at inference.
Da-Zhao Du, Shiyan Du, Jian Liu et al.· 0 citations
This work proposes OPTD, On-Policy Transition Distillation with consistency-guided adaptive compression, which consistently improves the quality--efficiency trade-off and attains the strongest overall quality-constrained AUP among the evaluated few-step baselines.
Xiaocheng Lu, Hualei Zhang, Shuhan Guo et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
