Metraipolis › archive › dc_56b5b60520ea

Token-Level Video Reinforcement Learning — arXiv:2610.01973

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.01973
License (as archived): CC0-1.0
Archived: 2026-10-06T19:15:13.290Z
SHA-256 of the archived text: 0cf4f66b262496e50d89c2736fa19650ae3687b79b33a1e5fd568efa9215df52
Stable URL of this page: https://metraipolis.com/documento/dc_56b5b60520ea

How to cite

Token-Level Video Reinforcement Learning — arXiv:2610.01973 — arXiv (en), https://arxiv.org/abs/2610.01973 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_56b5b60520ea (Metraipolis archive dc_56b5b60520ea, sha256 0cf4f66b262496e50d89c2736fa19650ae3687b79b33a1e5fd568efa9215df52)

BibTeX

@misc{dc56b5b60520ea,
  title = {Token-Level Video Reinforcement Learning — arXiv:2610.01973},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.01973},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_56b5b60520ea; sha256 0cf4f66b262496e50d89c2736fa19650ae3687b79b33a1e5fd568efa9215df52},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.01973. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reinforcement learning (RL) for video generation usually assigns one scalar reward to an entire sampled video. Yet a video is not uniformly flawed: some visual tokens may already satisfy the prompt, whereas others require correction. A scalar reward cannot localize errors, causing optimization to perturb satisfactory tokens while under-targeting the tokens that actually need to change. We introduce Token-Level Video Reinforcement Learning, TVRL, a framework that derives token-level credit from the reward being optimized. Our key insight is that the answer likelihood of a frozen vision-language model provides both signals: its outputs contribute to the video-level reward, while magnitudes of its video-input gradients reveal which generated video tokens most affect that score. We instantiate TVRL in Group Relative Policy Optimization by averaging prompt-derived question rewards into one group-relative advantage and using detached, question-conditioned token-credit maps to reweight dense denoising-transition log-probabilities inside the clipped policy ratio. On VBench-2.0, TVRL achieves an Overall score of 57.69, outperforming the base model by 3.60 points. TVRL also improves matched GRPO baselines across three SDE samplers (SAGE, Flow, and Dance) by 2.68--3.15 points and across four reward models (VideoAlign, VideoScore2, UnifiedReward2, and Qwen3.5-9B) by 1.33--3.15 points.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-10-02 in cs.CV).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.01973 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-06T17:37:36.869Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-10-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).