PROBE: Diagnosing Residual Concept Capacity in Erased Text-to-Video Diffusion Models
PROBE: 诊断 erased 文本到视频扩散模型中的残余概念容量
机构 * School of Artificial Intelligence and Automation, Huazhong University of Science and Technology(华中科技大学人工智能与自动化学院) ; Department of Computer Science and Engineering, University of Nevada(内华达大学计算机科学与工程系)
专题命中 视频扩散模型 :text-to-video(title,abstract);video diffusion(title);分类 cs.CV
AI总结 PROBE 通过多级评估框架揭示文本到视频扩散模型中被擦除概念的残余容量,发现所有测试方法留有可测量的残余能力,且其鲁棒性与干预深度相关,指出当前擦除方法仅实现输出级抑制而非表征移除。
Comments This preprint was posted after submission to IEEE Transactions