研究人员开发了 ZODS-RS，这是一个专为遥感图像中的零训练目标检测和分割设计的新型管道。该系统集成了 DINOv3 的密集特征和 SAM 风格的提议，可在无需特定任务训练数据的情况下生成水平边界框和实例掩码。ZODS-RS 在 FAIR1M 和 xView 等数据集上展示了改进的性能，尤其是在小目标和密集目标方面，并在无人机图像上显示出比现有方法（如 Grounded-SAM）显著的优势。 AI
影响 这种零训练方法可以简化遥感人工智能的部署，从而能够更快地适应新的平台和视角。
排序理由 该集群包含一篇 arXiv 论文，详细介绍了一种用于计算机视觉任务的新方法。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Remote-sensing and UAV applications need models that generalize across platforms and viewpoints without task-specific training. Yet training-free pipelines often falter on oriented geometry, scale/rotation variation, and crowded ports or airfields, and rarely unify detection and …
arXiv:2606.10769v1 Announce Type: new Abstract: Remote-sensing and UAV applications need models that generalize across platforms and viewpoints without task-specific training. Yet training-free pipelines often falter on oriented geometry, scale/rotation variation, and crowded por…
Remote-sensing and UAV applications need models that generalize across platforms and viewpoints without task-specific training. Yet training-free pipelines often falter on oriented geometry, scale/rotation variation, and crowded ports or airfields, and rarely unify detection and …