研究人员开发了SAM-Deep-EIoU，一种新颖的多目标跟踪方法，仅在基础跟踪器遇到不确定性时选择性地采用更强大的视频对象分割（VOS）模型。该方法无需训练，并将现有跟踪器视为黑盒，旨在提高在挑战性帧中的身份保持能力。在SportsMOT基准测试中，SAM-Deep-EIoU取得了86.8 HOTA分数的最先进性能。 AI
影响 通过智能地利用更强大的分割模型来提高对象跟踪的准确性，有可能改善体育分析和视频监控中的应用。
排序理由 这是一篇描述多目标跟踪新算法的研究论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.13033v1 Announce Type: new Abstract: Multi-object tracking has a heavy-tailed difficulty distribution: most frames are easy for a lightweight base tracker, while a small fraction are intrinsically hard. Video object segmentation (VOS) models can often preserve identity…
Multi-object tracking has a heavy-tailed difficulty distribution: most frames are easy for a lightweight base tracker, while a small fraction are intrinsically hard. Video object segmentation (VOS) models can often preserve identity through the hard frames where the base tracker …