arXiv · 机器人与计算机视觉· Ohad Rahamim, Dvir Samuel, Idan Schwartz, Gal Chechik·· 2 天前AI 评分55
All modalities are equal, but video is more equal: Closing the Cross-Attention Gap in Joint Video Generation
All modalities are equal, but video is more equal: Closing the Cross-Attention Gap in Joint Video Generation
AI 导读
论文提出 RecCAR(Reciprocal Cross-modal Attention Regularization)方法,针对联合视频与动作、音频生成的跨模态对应关系不对称问题,利用已建立的视频到模态的强对应作为固定参考,通过 KL 散度正则化对齐较弱的模态到视频对应。
来源:arXiv · 机器人与计算机视觉 · arxiv.org