Revisiting Feature Alignment for One-stage Object Detection
Yuntao Chen1,3 Chenxia Han Naiyan Wang2 Zhaoxiang Zhang1,3,4 1 University of Chinese Academy of Sciences 2 TuSimple 3 Center for Research on Intelligent Perception and Computing, CASIA 4Center for Excellence in Brain Science and Intelligence Technology, CAS {chenyuntao2016, {chenxiahan18,
Abstract
Recently, one-stage object detectors gain much attention due to their simplicity in practice. Its fully convolutional nature greatly reduces the difficulty of training and deployment compared with two-stage detectors which require NMS and sorting for the proposal stage. However, a fundamental issue lies in all one-stage detectors is the misalignment between anchor boxes and convolutional features, which significantly hinders the performance of one-stage detectors. In this work, we first reveal the deep connection between the widely used im2col operator and the RoIAlign operator. Guided by this illuminating observation, we propose a RoIConv operator which aligns the features and its corresponding anchors in one-stage detection in a principled way. We then design a fully convolutional AlignDet architecture which combines the flexibility of learned anchors and the preciseness of aligned features. Specifically, our AlignDet achieves a state-of-the-art mAP of 44.1 on the COCO test-dev with ResNeXt-101 backbone.
中文速览
单阶段目标检测器（one-stage detector）因为全卷积结构简洁高效而备受关注，但它存在一个根本性缺陷：卷积特征和对应锚框（anchor box）之间的空间错位，严重拖累了检测精度。论文作者发现，深度学习框架中常用的 im2col 算子本质上是 RoIAlign 的一种特例，两者在数学上高度统一，由此推导出一个名为 RoIConv 的新算子——它能根据每个锚框的实际位置和形状，精确计算采样偏移量，从而在不增加任何额外计算量的前提下，将特征与锚框严格对齐。基于 RoIConv，作者进一步构建了全卷积 AlignDet 检测框架，将"从数据中学习锚框分布"的灵活性与精准特征对齐结合在一起，在 COCO test-dev 上以 ResNeXt-101 骨干网络达到 44.1 mAP，比强基线 RetinaNet 提升 3.3 个点且速度不变。这项工作从理论层面厘清了一阶段检测器特征错位的本质，并给出了即插即用的解决方案，对推动高效精准目标检测具有重要意义。
原文 arXiv:1908.01570;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1908.01570v1