FreeAnchor: Learning to Match Anchors for Visual Object Detection

By Xiaosong Zhang, Fang Wan, Chang Liu, Rongrong Ji, Qixiang Ye. 2019

328 citations recorded by the source corpus.

Modern CNN-based object detectors assign anchors for ground-truth objects under the restriction of object-anchor Intersection-over-Unit (IoU). In this study, we propose a learning-to-match approach to break IoU restriction, allowing objects to match anchors in a flexible manner. Our approach, referred to as FreeAnchor, updates hand-crafted anchor assignment to "free" anchor matching by formulating detector training as a maximum likelihood estimation (MLE) procedure. FreeAnchor targets at learning features which best explain a class of objects in terms of both classification and localization. FreeAnchor is implemented by optimizing detection customized likelihood and can be fused with CNN-based detectors in a plug-and-play manner. Experiments on MS-COCO demonstrate that FreeAnchor consistently outperforms the counterparts with significant margins.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, background): “Thereafter, plenty of works are presented to boost its performance in different aspects, such as fusing context information from different layers [24, 12, 69], training from scratch [50, 73], introducing new loss function [33, 6], anchor refinement and matching [66, 67], architecture redesign [21, 2”
YOLOv4: Optimal Speed and Accuracy of Object Detection (2020, background): “FreeAnchor: Learning to match anchors for visual object detection [96] FreeAnchor ResNet-101 - 9.”
You Only Look One-level Feature (2021, methodology): “More importantly, the difference between the uniform matching and the learning-to-match methods is that: the learning-to-match methods, such as FreeAnchor [49] and PAA [16], adaptively separate anchors into positives and negatives according to the learning status, while uniform matching is ﬁxed and ”
Learning to Match Anchors for Visual Object Detection (2021, methodology): “The learning-to-match method was ﬁrst proposed in our NeurIPS 2019 paper [13] and is promoted theoretically and experimentally in this full version.”
Probabilistic Anchor Assignment with IoU Prediction for Object Detection (2020, methodology): “These works include selecting positive samples based on the detection-specific likelihood [42], the statistics of anchor IoUs [40] or the cleanness score of anchors [16,20].”
Probabilistic two-stage detection (2021, methodology): “While one-stage detectors have achieved competitive performance (Zhang et al., 2020b; Kim & Lee, 2020; Zhang et al., 2019; Li et al., 2020b; Zhu et al., 2020a), they usually rely on heavier separate classification and regression branches than two-stage models.”
Scale Match for Tiny Person Detection (2019, positive): “For adaptive FreeAnchor[29], we use same learning rate and backbone setting of Adaptive RetinaNet, and others are keep same as FreeAnchor’s default setting.”
RepPoints V2: Verification Meets Regression for Object Detection (2020, neutral): “9 FreeAnchor* [33] ResNeXt-101 24 47.”
Oriented RepPoints for Aerial Object Detection (2021, methodology): “Some recent samples assignment methods in general object detection, such as the ATSS [50], FreeAnchor [51], PAA [13] and OTA [6], employ a learning-tomatch optimization strategy [52] to choose the high-quality samples.”
VarifocalNet: An IoU-aware Dense Object Detector (2020, methodology): “While popular two-stage methods [3, 4] and multi-stage meth-ods [15] usually employ anchors to generate object proposals for downstream classiﬁcation and regression, anchor-based one-stage methods [6–8, 16, 17] directly classify and reﬁne anchor boxes without object proposal generation.”
CSPNet: A New Backbone that can Enhance Learning Capability of CNN (2019, background): “Recently, anchor-free based object detector [4, 43, 13, 14, 40] has become main-stream object detection system.”
Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (2020, background): “Some researches (Zhang et al. 2019; Song, Liu, and Wang 2020) have reported that the discriminative features required to localize objects are not evenly distributed on GT, especially for objects with a wide variety of orientations and aspect ratios.”
Effective Fusion Factor in FPN for Tiny Object Detection (2020, neutral): “13 FreeAnchor‡ [35] 89.”
Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training (2020, neutral): “0 FreeAnchor [50] ResNet-101 41.”
Prototype Mixture Models for Few-shot Semantic Segmentation (2020, background): “However, annotating object masks for large-scale datasets is laborious, expensive, and can be impractical [9,10,11].”
TOOD: Task-aligned One-stage Object Detection (2021, methodology): “As discussed in [31, 32], training sample assignment is crucial to the training of object detectors.”