Ruopeng Gao, Yukun Huang, Jiarui Hu, Shuai Wang, Wenbo Hu, Ying Shan, Limin Wang · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.29989 · cited by 0 works (OpenAlex) · open-access PDF
Los modelos de difusión de vídeo autoregresivos generan videos chunk por chunk, sin embargo, la producción de continuos plausibles no garantiza que capturen cómo evolucionan las escenas. Bajo la fuerza del maestro, la fuerte redundancia temporal puede permitir que la continuidad visual local sirva como un paso hacia atrás para el denosamiento de corriente, reduciendo la presión para modelar la evolución visual. Introducimos Future Chunk Prediction (FCP), un objetivo auxiliar optimizado conjuntamente con el denosamiento forzado por el maestro que predice fragmentos temporalmente separados sin acceso al contexto visual que interviene. Atrayendo la supervisión directamente de los vídeos de formación, el FCP no requiere un codificador visual pre-entrenado adicional y deja inalterada la inferencia autoregresiva. Nuestro método logra una plaus
Machine translation (Meta M2M100); the English original follows.
Autoregressive video diffusion models generate videos chunk by chunk, yet producing plausible continuations does not guarantee that they capture how scenes evolve. Under teacher forcing, strong temporal redundancy may allow local visual continuity to serve as a shortcut for current-chunk denoising, reducing the pressure to model visual evolution. We introduce Future Chunk Prediction (FCP), an auxiliary objective jointly optimized with teacher-forced denoising that predicts temporally separated chunks without access to the intervening visual context. Drawing supervision directly from training videos, FCP requires no additional pretrained visual encoder and leaves autoregressive inference unchanged. Our method achieves superior physical plausibility over standard teacher-forced autoregressive training with a lower overall training compute budget, highlighting the effectiveness and efficiency of future-chunk supervision for modeling visual evolution. Intriguingly, our visualizations suggest that FCP fosters a more structured representation space, which may be valuable for building autoregressive world models.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Autoregressive model, Computer science, Redundancy (engineering), Inference, Artificial intelligence, Encoder
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.29989, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange