研究人员开发了 MDS-DETR，这是一种新颖的目标检测模型，它改进了 DEtection TRansformer (DETR) 架构。MDS-DETR 通过在单个解码器中集成一对一和一对多标签分配，解决了 DETR 收敛慢和召回率低的问题。这是通过掩码重复抑制器 (MDS) 实现的，该抑制器过滤冗余预测，从而实现更高效、更准确的目标检测。 AI
影响 MDS-DETR 为目标检测任务提供了改进的训练效率和准确性，可能使计算机视觉应用受益。
排序理由 该集群包含一篇详细介绍目标检测新模型架构的研究论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2605.23507v1 Announce Type: new Abstract: The DEtection TRansformer (DETR) is a powerful end-to-end object detector, yet its one-to-one matching strategy suffers from slow convergence and low recall. A common approach to address this issue is to use one-to-many label assign…
The DEtection TRansformer (DETR) is a powerful end-to-end object detector, yet its one-to-one matching strategy suffers from slow convergence and low recall. A common approach to address this issue is to use one-to-many label assignment to provide more positive samples. However, …