arXiv 2610.145532026-10-14cs.CV

用于鲁棒少步自回归视频生成的对抗梯度强制

Adversarial Gradient Forcing for Robust Few-Step Autoregressive Video Generation

Zhejiang University（浙江大学）；Alibaba Group（阿里巴巴集团）；Nanjing University（南京大学）

Chenxi Xia, Yixuan Xu, Shan Liu, Jiapeng Wang, Maohua Li, Hanlin Tang, Kan Liu, Tao Lan, Weiyi Lu, Xinggao Liu

备注：: Project page: https://wuli-art.github.io/Adversarial-Gradient-Forcing/ ; Code: https://github.com/wuli-art/Adversarial-Gradient-Forcing

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对自回归视频生成的局部历史鲁棒性差距，提出AGF框架，通过联合训练原始与扰动历史提升鲁棒性，在60秒视频生成获VBench最高分，4分钟视频生成视觉质量最优，仅需5秒回推训练。

摘要：

自回归视频生成模型按顺序生成视频，每个新块都以先前生成的块为条件。通过在模型自身的回推（rollouts）上进行训练，自强制（Self Forcing）及其后续方法大幅缓解了由训练与推理阶段历史分布不匹配导致的暴露偏差。然而，这些模型在长时序生成过程中仍存在误差累积问题。我们发现了一个关键的潜在局限：模型对自身生成历史中的误差仍高度敏感，允许微小偏差通过后续生成被放大，我们将这一未被充分研究的问题称为局部历史鲁棒性差距。为解决该差距，我们提出对抗梯度强制（Adversarial Gradient Forcing，AGF）框架，旨在提升少步自回归视频模型对自身生成历史中的扰动和累积误差的鲁棒性。AGF利用未来块预测的分布匹配梯度，选择针对自生成历史的对抗扰动，随后在原始和扰动后的历史上联合训练生成器，以提升未来生成的鲁棒性。大量实验表明，AGF在60秒视频生成任务中取得了最高的VBench总得分，在4分钟视频生成任务中实现了最佳视觉质量。值得注意的是，AGF仅需5秒的回推即可训练，同时实现了如此长时序的鲁棒性。

英文摘要：

Autoregressive video generation models generate videos sequentially, with each new block conditioned on previously generated blocks. By training on the model's own rollouts, Self Forcing and its follow-up methods substantially mitigate exposure bias caused by the mismatch between training- and inference-time history distributions. However, these models still suffer from error accumulation during long-horizon generation. We identify a key underlying limitation: the model remains highly sensitive to errors in its generated history, allowing small deviations to be amplified through subsequent generations. We refer to this underexplored problem as the local history robustness gap. To address this gap, we introduce Adversarial Gradient Forcing (AGF), a framework designed to improve the robustness of few-step autoregressive video models to perturbations and accumulated errors in their own generated histories. AGF uses distribution matching gradients from future block predictions to select adversarial perturbations of self-generated histories. It then jointly trains the generator on the original and perturbed histories to improve the robustness of future generation. Extensive experiments show that AGF achieves the highest VBench Total score on 60-second video generation and the best visual quality on four-minute video generation. Notably, AGF achieves such long-horizon robustness while requiring only 5-second rollouts for training.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。