arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2304.07527cs.CV
收藏

Align-DETR:利用对齐损失增强端到端目标检测

Align-DETR: Enhancing End-to-end Object Detection with Aligned Loss

Beihang University(北京航空航天大学);Megvii Inc.(旷视科技有限公司)

Zhi Cai, Songtao Liu, Guodong Wang, Zheng Ge, Xiangyu Zhang, Di Huang

AI总结:

本文提出Align-DETR,通过引入Align Loss解决分类-回归错位与跨层目标错位问题,结合多对一匹配监督,显著提升了DETR的收敛速度与检测性能。

摘要:

DETR通过将目标检测任务构建为集合预测问题,建立了一个简单的端到端流水线,展现出广阔的前景。尽管取得了显著进展,本文仍识别出模型内部存在的两种关键错位:分类-回归错位和跨层目标错位。这两个问题均会阻碍DETR的收敛并降低其整体性能。为同时解决这两个问题,我们引入了一种名为Align Loss的新型损失函数,旨在消除两项任务之间的差异。Align Loss通过联合质量指标引导DETR的优化过程,强化了分类与回归之间的联系。此外,它还引入了一个指数降权项,以实现正样本到负样本的平滑过渡。Align-DETR还采用多对一匹配来监督中间层,类似于H-DETR的设计,从而增强了模型对不稳定性的鲁棒性。我们进行了大量实验,取得了极具竞争力的结果。值得注意的是,在采用ResNet-50骨干网络的H-DETR基线上,我们的方法实现了49.3%(+0.6)的AP。它还创造了新的最先进性能,在1x设置下达到50.5% AP,在2x设置下达到51.7% AP,超越了多个强大的竞争对手。代码发布于https://github.com/FelixCaae/AlignDETR。

英文摘要:

DETR has set up a simple end-to-end pipeline for object detection by formulating this task as a set prediction problem, showing promising potential. Despite its notable advancements, this paper identifies two key forms of misalignment within the model: classification-regression misalignment and cross-layer target misalignment. Both issues impede DETR's convergence and degrade its overall performance. To tackle both issues simultaneously, we introduce a novel loss function, termed as Align Loss, designed to resolve the discrepancy between the two tasks. Align Loss guides the optimization of DETR through a joint quality metric, strengthening the connection between classification and regression. Furthermore, it incorporates an exponential down-weighting term to facilitate a smooth transition from positive to negative samples. Align-DETR also employs many-to-one matching for supervision of intermediate layers, akin to the design of H-DETR, which enhances robustness against instability. We conducted extensive experiments, yielding highly competitive results. Notably, our method achieves a 49.3% (+0.6) AP on the H-DETR baseline with the ResNet-50 backbone. It also sets a new state-of-the-art performance, reaching 50.5% AP in the 1x setting and 51.7% AP in the 2x setting, surpassing several strong competitors. Our code is available at https://github.com/FelixCaae/AlignDETR.

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。

↑