The source code includes training and inference procedures for the proposed method of the paper submitted to the Thirty-Sixth AAAI Conference on Artificial Intelligence (AAAI 2022) with title "Shape-Adaptive Selection and Measurement for Oriented Object Detection" (ID: 2171).
The the effectiveness of the proposed method is verified on two baseline methods. Corresponding source code and configurations reside in following two sub-directories:
sasm_reppoints
: the implementation and verification using RepPoints as baseline;sasm_s2anet
: the implementation and verification using s2anet(https://ieeexplore.ieee.org/document/9377550) as baseline.We provide only the source code related to the proposed method in the sub-directories so that reviewers can check them quickly and conveniently.
Please refer to the README.md
file in each sub-directory for the detailed instructions of usage.
Method | Assignment | Reg. Loss | Tricks | mAP |
---|---|---|---|---|
RepPoints | MaxIoU | GIoU | - | 70.46 |
RepPoints | SASM | BCLoss + GIoU | - | 74.27 |
RepPoints | SASM | BCLoss + GIoU | MS training | 77.19 |
s2anet | SASM | Smooth L1 | MS training | 79.17 |
1、https://github.com/open-mmlab/mmdetection
2、https://github.com/LiWentomng/OrientedRepPoints
3、https://github.com/SDL-GuoZonghao/BeyondBoundingBox