MetaAnchor: Learning to Detect Objects with Customized Anchors

By Tong Yang, Xiangyu Zhang, Wenqiang Zhang, Jian‐Yuan Sun. 2018

151 citations recorded by the source corpus.

We propose a novel and flexible anchor mechanism named MetaAnchor for object detection frameworks. Unlike many previous detectors model anchors via a predefined manner, in MetaAnchor anchor functions could be dynamically generated from the arbitrary customized prior boxes. Taking advantage of weight prediction, MetaAnchor is able to work with most of the anchor-based object detection systems such as RetinaNet. Compared with the predefined anchor scheme, we empirically find that MetaAnchor is more robust to anchor settings and bounding box distributions; in addition, it also shows the potential on the transfer task. Our experiment on COCO detection task shows MetaAnchor consistently outperforms the counterparts in various scenarios.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Meta & Continual Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Meta & Continual Learning

Citations in the corpus

Dynamic Neural Networks: A Survey (2021, background): “Image Sa Object detection (face [40], [203], [204], facial point [205], pedestrian [206], general [33], [207], [208], [209], [210]) Image segmentation [107], [211], [212], Super resolution [213], Style transfer [214], Coarse-to-fine classification [215]”
DetNAS: Backbone Search for Object Detection (2019, background): “MetaAnchor [28] proposes a dynamic anchor mechanism to boost the performance for anchor-based object detectors.”
Learning Data Augmentation Strategies for Object Detection (2019, background): “simple and well known techniques for improving object detection performance [44, 16].”
Region Proposal by Guided Anchoring (2019, neutral): “MetaAnchor [32] introduces meta-learning to anchor generation.”
Detection in Crowded Scenes: One Proposal, Multiple Predictions (2020, background): “…the paradigm of proposal-based object detectors mainly involves two steps: the ﬁrst step is proposal box generation , which could be achieved by Selective Search [12, 11], predeﬁned/learnable anchors [29, 28, 24, 21, 41, 39, 45] or Region Proposal Networks (RPNs) [29, 20, 13, 4, 2], etc.”
DetNAS: Neural Architecture Search on Object Detection (2019, background): “MetaAnchor [25] proposes a dynamic anchor mechanism to boost the performance for anchor-based object detectors.”
FoveaBox: Beyound Anchor-Based Object Detection (2019, background): “(b) One design choice based on a particular dataset is not always applicable to other applications, which harms the generality [12].”
OTA: Optimal Transport Assignment for Object Detection (2021, background): “MetaAnchor [44] proposes an anchor generation function to learn dynamic anchors from the arbitrary customized prior boxes.”
Recent Advances in Deep Learning for Object Detection (2019, unknown): “[93] modeled anchors as func-”
Imbalance Problems in Object Detection: A Review (2019, methodology): “The Meta Anchor [133] method attempts to find an optimal set of aspect ratios and scales for anchors.”
Probabilistic Anchor Assignment with IoU Prediction for Object Detection (2020, background): “MetaAnchor [37] predicts the parameters of the anchor functions (the last convolutional layers of detection heads) dynamically and takes anchor shapes as an argument, which provides the ability to change anchors in training and testing.”
FoveaBox: Beyond Anchor-based Object Detector (2019, background): “In MetaAnchor [46], anchor functions are dynamically generated from the arbitrary customized prior boxes.”
FreeAnchor: Learning to Match Anchors for Visual Object Detection (2019, methodology): “MetaAnchor [16] approach learns to produce anchors from the arbitrary customized prior boxes with a sub-network.”
Meta-SR: A Magnification-Arbitrary Network for Super-Resolution (2019, background): “[30] proposed a novel and flexible anchor mechanism for object detection.”
MetaPruning: Meta Learning for Automatic Neural Network Channel Pruning (2019, background): “Recent works also applies meta learning on various tasks and achieves state-ofthe-art results in detection [51], super-resolution with arbitrary magnification [27] and instance segmentation [26].”
An Asymptotically Optimal Multi-Armed Bandit Algorithm and Hyperparameter Optimization (2020, methodology): “In addition, since the training procedures of MetaAnchor (Yang et al., 2018) and Zhong’s method (Zhong et al.”