研究人员开发了一种新颖的无监督跨域昼夜重识别框架。该方法采用两阶段训练策略，结合了提示学习和基于原型的表示学习。通过利用视觉语言模型生成文本提示，并将视觉特征与这些提示对齐，系统在没有手动标注的情况下建立了昼夜场景之间的身份对应关系。实验表明，该方法取得了与最先进的完全监督方法相当的性能。 AI
影响 引入了一种新颖的无监督重识别方法，有望减少监控和图像检索系统中对昂贵手动标注的依赖。
排序理由 该集群包含一篇详细介绍计算机视觉任务新方法的学术论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.12258v1 Announce Type: new Abstract: Cross-domain day-night re-identification (ReID) is fundamentally challenged by the substantial visual appearance discrepancies between daytime and nighttime scenes. Existing fully supervised methods rely heavily on labor-intensive a…
Cross-domain day-night re-identification (ReID) is fundamentally challenged by the substantial visual appearance discrepancies between daytime and nighttime scenes. Existing fully supervised methods rely heavily on labor-intensive annotations, which are costly and exhibit limited…