arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2307.08353cs.CV
收藏

Box-DETR:理解并框定条件空间查询

Box-DETR: Understanding and Boxing Conditional Spatial Queries

Huazhong University of Science and Technology(华中科技大学)

Wenze Liu, Hao Lu, Yuliang Liu, Zhiguo Cao

AI总结:

针对 DAB-DETR 交叉注意力仅使用框中心导致尺寸信息缺失的问题,本文提出 Box Agent,将完整框压缩为各头代理点作为参考,从而加快收敛并提升检测精度。

摘要:

条件空间查询最近被引入 DEtection TRansformer(DETR)以加速收敛。在 DAB-DETR 中,此类查询在每个解码器阶段由所谓的条件线性投影进行调制,旨在搜索感兴趣的位置,例如框的四个端点。每个解码器阶段通过预测锚框偏移量来逐步更新框,而在交叉注意力中,只有框中心被作为参考点告知。然而,仅使用框中心会使当前阶段无法获知前一个框的宽度和高度,从而阻碍对偏移量的准确预测。我们认为,在交叉注意力中显式使用完整框信息至关重要。在本文中,我们提出 Box Agent,将框压缩为特定于注意力头的代理点。通过在每个注意力头中用代理点替代框中心作为参考点,条件交叉注意力能够在考虑前一个框完整范围的情况下,从更合理的起点搜索位置,而不是总是从前一个框中心出发。这显著减轻了条件线性投影的负担。实验结果表明,box agent 不仅带来更快的收敛,还提升了检测性能;例如,在 DAB-DETR 基础上使用 ResNet-50 时,我们的单尺度模型达到 44.2 AP。我们的 Box Agent 只需对代码进行少量修改,并且计算开销可忽略不计。代码可在 https://github.com/tiny-smart/box-detr 获取。

英文摘要:

Conditional spatial queries are recently introduced into DEtection TRansformer (DETR) to accelerate convergence. In DAB-DETR, such queries are modulated by the so-called conditional linear projection at each decoder stage, aiming to search for positions of interest such as the four extremities of the box. Each decoder stage progressively updates the box by predicting the anchor box offsets, while in cross-attention only the box center is informed as the reference point. The use of only box center, however, leaves the width and height of the previous box unknown to the current stage, which hinders accurate prediction of offsets. We argue that the explicit use of the entire box information in cross-attention matters. In this work, we propose Box Agent to condense the box into head-specific agent points. By replacing the box center with the agent point as the reference point in each head, the conditional cross-attention can search for positions from a more reasonable starting point by considering the full scope of the previous box, rather than always from the previous box center. This significantly reduces the burden of the conditional linear projection. Experimental results show that the box agent leads to not only faster convergence but also improved detection performance, e.g., our single-scale model achieves $44.2$ AP with ResNet-50 based on DAB-DETR. Our Box Agent requires minor modifications to the code and has negligible computational workload. Code is available at https://github.com/tiny-smart/box-detr.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑