Author name: Zhen Qu, Xian Tao, Xiaoyi Bao, Dingrong Wang, ShiChen Qu, Zhengtao Zhang, Xingang Wang

arXiv:2603.01305v2 Announce Type: replace
Abstract: Large multimodal models (LMMs) exhibit strong task generalization capabilities, offering new opportunities for zero-shot visual anomaly segmentation (ZSAS). However, existing LMM-based segmentation a…