MetaAnchor: Learning to Detect Objects with Customized Anchors
Abstract
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.
Cite
Text
Yang et al. "MetaAnchor: Learning to Detect Objects with Customized Anchors." Neural Information Processing Systems, 2018.Markdown
[Yang et al. "MetaAnchor: Learning to Detect Objects with Customized Anchors." Neural Information Processing Systems, 2018.](https://mlanthology.org/neurips/2018/yang2018neurips-metaanchor/)BibTeX
@inproceedings{yang2018neurips-metaanchor,
title = {{MetaAnchor: Learning to Detect Objects with Customized Anchors}},
author = {Yang, Tong and Zhang, Xiangyu and Li, Zeming and Zhang, Wenqiang and Sun, Jian},
booktitle = {Neural Information Processing Systems},
year = {2018},
pages = {320-330},
url = {https://mlanthology.org/neurips/2018/yang2018neurips-metaanchor/}
}