试着改述这一段英语This paper investigates a phenomenon where query- based object detectors mispredict at the last decoding stage while predicting correctly at an intermediate stage. We review the training process and attribute the overlooked phenomenon to two limitations: lack of training emphasis and cascading errors from decoding sequence. We design and present Selective Query Recollection (SQR), a simple and effective training strategy for query-based object detec- tors. It cumulatively collects intermediate queries as decod- ing stages go deeper and selectively forwards the queries to the downstream stages aside from the sequential struc- ture. Such-wise, SQR places training emphasis on later stages and allows later stages to work with intermediate queries from earlier stages directly. SQR can be easily plugged into various query-based object detectors and sig- nificantly enhances their performance while leaving the in- ference pipeline unchanged. As a result, we apply SQR on Adamixer, DAB-DETR, and Deformable-DETR across various settings (backbone, number of queries, schedule) and consistently brings 1.4 ∼ 2.8 AP improvement.

这篇论文研究了一个现象，即在预测中间阶段正确的情况下，基于查询的物体检测器在最后的解码阶段误判。我们回顾了训练过程，并将这种被忽视的现象归因于两个限制：缺乏训练重点和解码序列中的级联错误。我们设计并提出了选择性查询回收（SQR），这是一种简单而有效的基于查询的物体检测器训练策略。随着解码阶段的深入，SQR累积收集中间查询，并选择性地将查询转发到下游阶段，除了顺序结构之外。这样，SQR将训练重点放在后期阶段，并允许后期阶段直接使用来自早期阶段的中间查询。SQR可以轻松地插入各种基于查询的物体检测器，并显著提高它们的性能，同时不改变推理流程。因此，我们在各种设置（骨干网络，查询数量，时间表）上将SQR应用于Adamixer，DAB-DETR和Deformable-DETR，并始终带来1.4〜2.8 AP的改进。