arXiv cs.AI· Hongyu Cao, Guangfu Guo, Jinghan Zhang, Dakshak Keerthi Chandra, Ying Zhang, Kunpeng Liu·· 4 小时前AI 评分34
GENIQ：用文生图模型自身生成图像实现皮肤病灶分割
Segment What It Can Draw: Querying a Text-to-Image Generator with Its Own Images for Skin Lesion Segmentation
AI 导读
一种名为 GENIQ 的免训练方法利用文生图扩散模型自身生成的健康皮肤与病灶图像构建概念方向，再结合无标注皮肤镜图像中最像皮肤和病灶的图块进行修正，完成无标注皮肤病灶分割。方法无需推理时标注或额外模型，在三个皮肤镜测试集上取得 0.85–0.88 的 Dice 分数，并在全部三个测试集上优于零样本和基于词注意力的方法。
正文 · 原文
Abstract:Skin lesion segmentation without annotated masks requires the target to be specified through other cues. Existing approaches rely on dermoscopy-specific rules, annotation-derived model priors, or user-selected regions. A general text-to-image diffusion transformer offers a text interface to the target, but its concept-word attention is unreliable on dermoscopy images: in the model's attention-output space, the word-derived skin-to-lesion direction is poorly aligned with the direction separating real skin from real lesions. We show that the direction derived from generated healthy-skin and lesion images is substantially better aligned. Based on this finding, we propose GENIQ, a training-free method that constructs initial skin and lesion prototypes from generated images and refines them using the most skin-like and lesion-like patches from unlabelled dermoscopy images. It scores test patches along the refined direction and applies a gated threshold to suppress contextual spill-over around small lesions. Without inference-time annotations or additional models, GENIQ achieves Dice scores of 0.85-0.88 on three dermoscopy test sets. It outperforms zero-shot and word-based attention methods on all three sets, and a hand-designed dermoscopy pipeline in Dice and IoU on two.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| ACM classes: | I.4.6; I.2.10 |
| Cite as: | arXiv:2611.07061 [cs.CV] |
| (or arXiv:2611.07061v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2611.07061 arXiv-issued DOI via DataCite |
Submission history
From: Hongyu Cao [view email]
[v1]
Wed, 4 Nov 2026 19:20:08 UTC (643 KB)
来源：arXiv cs.AI · arxiv.org