ORCA: Hunting Compositional Failures in Text-to-Image Diffusion

作者：Arshia Hemmat, Amirhossein Vahidi, Amitis Shidani, Mohammad Vali Sanian, Hesam Asadollahzadeh, Aryan Yazdan Parast, Mohammad Lotfollahi · 单位：Wellcome Sanger Institute, Cambridge Stem Cell Institute, University of Cambridge, Cambridge Centre for AI in Medicine, University of Cambridge, University of Oxford, Department of Computer Science, University of Helsinki, Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Department of Medicine, University of Cambridge · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-08