Metraipolis › archive › dc_339aadca02ad
SGF+: Decoupling Gradient Flows for Autoregressive Video Generation — arXiv:2610.10429
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.10429
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T12:48:07.415Z
- SHA-256 of the archived text
17ab48836a24cfda1303867d700300500a4de6c1cb15f3ad8358cb040a87b16b- Stable URL of this page
https://metraipolis.com/documento/dc_339aadca02ad
How to cite
SGF+: Decoupling Gradient Flows for Autoregressive Video Generation — arXiv:2610.10429 — arXiv (en), https://arxiv.org/abs/2610.10429 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_339aadca02ad (Metraipolis archive dc_339aadca02ad, sha256 17ab48836a24cfda1303867d700300500a4de6c1cb15f3ad8358cb040a87b16b)
BibTeX
@misc{dc339aadca02ad,
title = {SGF+: Decoupling Gradient Flows for Autoregressive Video Generation — arXiv:2610.10429},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.10429},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_339aadca02ad; sha256 17ab48836a24cfda1303867d700300500a4de6c1cb15f3ad8358cb040a87b16b},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.10429. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Autoregressive video generation requires denoising the current frames while writing their key-value representations as context for future predictions. However, these two roles typically share parameters, and we find that their gradients exhibit distinct patterns and systematic negative alignment, hindering the joint optimization of visual quality and temporal consistency. We introduce Self Gradient Forcing Plus (SGF+), which assigns separate parameters to context writing and denoising while preserving their interaction through causal attention. Both roles are jointly optimized using the original generation objective without auxiliary losses, with context writing supervised through its contribution to future predictions. This simple change improves visual quality and long-horizon consistency over the evaluated baselines in both framewise and chunkwise generation, without additional video training data or a longer training horizon. Trained on only 5s rollouts, SGF+ supports continuous generation for up to 24 hours without long-video fine-tuning. These results highlight role-specific parameterization as an effective design principle for high-quality autoregressive video generation and native long-horizon extrapolation. Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV). Pagina del lavoro: https://arxiv.org/abs/2610.10429 — 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 2026-10-08T12:18:15.497Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).