Revisiting Feature Alignment for One-stage Object Detection

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.

References in corpus (11)

• YOLOv3: An Incremental Improvement
• DSSD : Deconvolutional Single Shot Detector
• Focal Loss for Dense Object Detection
• FCOS: Fully Convolutional One-Stage Object Detection
• DenseBox: Unifying Landmark Localization with End to End Object Detection
• Deformable Convolutional Networks
• CenterNet: Keypoint Triplets for Object Detection
• CornerNet: Detecting Objects as Paired Keypoints
• RepPoints: Point Set Representation for Object Detection
• Bottom-up Object Detection by Grouping Extreme and Center Points
• Region Proposal by Guided Anchoring

Cited by in corpus (16)

• Probabilistic two-stage detection
• Attentional Feature Refinement and Alignment Network for Aircraft Detection in SAR Imagery
• PDNet: Toward Better One-Stage Object Detection With Prediction Decoupling
• CentripetalNet: Pursuing High-quality Keypoint Pairs for Object Detection
• Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training
• Corner Proposal Network for Anchor-free, Two-stage Object Detection
• Align Deep Features for Oriented Object Detection
• IoU-aware Single-stage Object Detector for Accurate Localization
• SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection
• MimicDet: Bridging the Gap Between One-Stage and Two-Stage Object Detection
• Instance Localization for Self-supervised Detection Pretraining
• AlignSeg: Feature-Aligned Segmentation Networks
• SaccadeNet: A Fast and Accurate Object Detector
• Disentangle Your Dense Object Detector
• M3DSSD: Monocular 3D Single Stage Object Detector
• EHSOD: CAM-Guided End-to-end Hybrid-Supervised Object Detection with Cascade Refinement