Skip to content

Latest commit

 

History

History
63 lines (47 loc) · 10.2 KB

README.md

File metadata and controls

63 lines (47 loc) · 10.2 KB

Region Proposal by Guided Anchoring

Abstract

Region anchors are the cornerstone of modern object detection techniques. State-of-the-art detectors mostly rely on a dense anchoring scheme, where anchors are sampled uniformly over the spatial domain with a predefined set of scales and aspect ratios. In this paper, we revisit this foundational stage. Our study shows that it can be done much more effectively and efficiently. Specifically, we present an alternative scheme, named Guided Anchoring, which leverages semantic features to guide the anchoring. The proposed method jointly predicts the locations where the center of objects of interest are likely to exist as well as the scales and aspect ratios at different locations. On top of predicted anchor shapes, we mitigate the feature inconsistency with a feature adaption module. We also study the use of high-quality proposals to improve detection performance. The anchoring scheme can be seamlessly integrated into proposal methods and detectors. With Guided Anchoring, we achieve 9.1% higher recall on MS COCO with 90% fewer anchors than the RPN baseline. We also adopt Guided Anchoring in Fast R-CNN, Faster R-CNN and RetinaNet, respectively improving the detection mAP by 2.2%, 2.7% and 1.2%.

Citation

We provide config files to reproduce the results in the CVPR 2019 paper for Region Proposal by Guided Anchoring.

@inproceedings{wang2019region,
    title={Region Proposal by Guided Anchoring},
    author={Jiaqi Wang and Kai Chen and Shuo Yang and Chen Change Loy and Dahua Lin},
    booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
    year={2019}
}

Results and Models

The results on COCO 2017 val is shown in the below table. (results on test-dev are usually slightly higher than val).

Method Backbone Style Lr schd Mem (GB) Inf time (fps) AR 1000 Config Download
GA-RPN R-50-FPN caffe 1x 5.3 15.8 68.4 config model | log
GA-RPN R-101-FPN caffe 1x 7.3 13.0 69.5 config model | log
GA-RPN X-101-32x4d-FPN pytorch 1x 8.5 10.0 70.6 config model | log
GA-RPN X-101-64x4d-FPN pytorch 1x 7.1 7.5 71.2 config model | log
Method Backbone Style Lr schd Mem (GB) Inf time (fps) box AP Config Download
GA-Faster RCNN R-50-FPN caffe 1x 5.5 39.6 config model | log
GA-Faster RCNN R-101-FPN caffe 1x 7.5 41.5 config model | log
GA-Faster RCNN X-101-32x4d-FPN pytorch 1x 8.7 9.7 43.0 config model | log
GA-Faster RCNN X-101-64x4d-FPN pytorch 1x 11.8 7.3 43.9 config model | log
GA-RetinaNet R-50-FPN caffe 1x 3.5 16.8 36.9 config model | log
GA-RetinaNet R-101-FPN caffe 1x 5.5 12.9 39.0 config model | log
GA-RetinaNet X-101-32x4d-FPN pytorch 1x 6.9 10.6 40.5 config model | log
GA-RetinaNet X-101-64x4d-FPN pytorch 1x 9.9 7.7 41.3 config model | log
  • In the Guided Anchoring paper, score_thr is set to 0.001 in Fast/Faster RCNN and 0.05 in RetinaNet for both baselines and Guided Anchoring.

  • Performance on COCO test-dev benchmark are shown as follows.

Method Backbone Style Lr schd Aug Train Score thr AP AP_50 AP_75 AP_small AP_medium AP_large Download
GA-Faster RCNN R-101-FPN caffe 1x F 0.05
GA-Faster RCNN R-101-FPN caffe 1x F 0.001
GA-RetinaNet R-101-FPN caffe 1x F 0.05
GA-RetinaNet R-101-FPN caffe 2x T 0.05