PRBNet PyTorch
This is the reference PyTorch implementation for training and testing single-shot object detection and oriented bounding boxes models using the method described in
Parallel Residual Bi-Fusion Feature Pyramid Network for Accurate Single-Shot Object Detection
Ping-Yang, Chen, Ming-Ching Chang, Jun-Wei Hsieh, and Yong-Sheng Chen
Performance
MS COCO
P5 Model
Model | Test Size | APtest | AP50test | AP75test | APstest | FPS |
---|---|---|---|---|---|---|
YOLOv7-x | 640 | 53.1% | 71.2% | 57.8% | 33.8% | 114 |
PRB-FPN-CSP | 640 | 51.8% | 70.0% | 56.7% | 32.6% | 113 |
PRB-FPN-MSP | 640 | 53.3% | 71.1% | 58.3% | 34.1% | 94 |
PRB-FPN-ELAN | 640 | 52.5% | 70.4% | 57.2% | 33.4% | 70 |
P6 Model
Model | Test Size | APtest | AP50test | AP75test | FPS |
---|---|---|---|---|---|
YOLOv7-E6E | 1280 | 56.8% | 74.4% | 62.1% | 36 |
PRB-FPN6 | 1280 | 56.9% | 74.1% | 62.3% | 31 |
PRB-FPN6-MSP | 1280 | 57.2% | 74.5% | 62.5% | 27 |
If you find our work useful in your research please consider citing our paper:
@ARTICLE{9603994,
author={Chen, Ping-Yang and Chang, Ming-Ching and Hsieh, Jun-Wei and Chen, Yong-Sheng},
journal={IEEE Transactions on Image Processing},
title={Parallel Residual Bi-Fusion Feature Pyramid Network for Accurate Single-Shot Object Detection},
year={2021},
volume={30},
number={},
pages={9099-9111},
doi={10.1109/TIP.2021.3118953}}
If you find the backbone also well-done in your research, please consider citing the CSPNet. Most of the credit goes to Dr. Wang:
@inproceedings{wang2020cspnet,
title={{CSPNet}: A New Backbone That Can Enhance Learning Capability of {CNN}},
author={Wang, Chien-Yao and Mark Liao, Hong-Yuan and Wu, Yueh-Hua and Chen, Ping-Yang and Hsieh, Jun-Wei and Yeh, I-Hau},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
pages={390--391},
year={2020}
}
Acknowledgement
Without the guidance of Dr. Mark Liao and a discussion with Dr. Wang, PRBNet would not have been published quickly in TIP and open-sourced to the community. Many of the code is borrowed from YOLOv4, YOLOv5_obb, and YOLOv7. Many thanks for their fantastic work: