Contact us with tingtingliang@pku.edu.cn, wyt@pku.edu.cn.
This project provides an implementation for our CVPR2021 paper "OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection" on PyTorch. The search code is coming soon.
If you use our code/model/data, please cite our paper
@inproceedings{liang2021opanas,
title={Opanas: One-shot path aggregation network architecture search for object detection},
author={Liang, Tingting and Wang, Yongtao and Tang, Zhi and Hu, Guosheng and Ling, Haibin},
booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
pages={10195--10203},
year={2021}
}The project is only free for academic research purposes, but needs authorization for commerce. For commerce permission, please contact wyt@pku.edu.cn.
Results on COCO. Note that our Faster R-CNN uses smooth L1 loss following the original paper.
| Method | Backbone | Lr schd | box AP (val) | box AP (test-dev) | Download |
|---|---|---|---|---|---|
| Faster R-CNN | R-50 | 1x | 39.6 | 40.1 | model |
| Cascade R-CNN | R2-101 | 2x | 51.8 | 52.2 | model |
Please refer to install.md for installation and dataset preparation. You need to install mmdetection (version 2.4.0 with mmcv 1.1.6) firstly. More guidance can be found from mmdeteion.
Please see getting_started.md for the basic usage of MMDetection. We use 8 GPUs (32GB V100) to train our detector, you can adjust the batch size in configs by yourselves.
./tools/dist_train.sh configs/opanas/faster_rcnn_r50_opa_fpn_112_sml1_coco.py 8 ./tools/dist_train.sh configs/opanas/cascade_rcnn_2r101_dcn_opa_fpn_160_2x_ms_coco.py 8
./tools/dist_test.sh configs/opanas/faster_rcnn_r50_opa_fpn_112_sml1_coco.py /path/to/your/save_dir/faster_opa_396.pth 8 --eval bbox ./tools/dist_test.sh configs/opanas/cascade_rcnn_2r101_dcn_opa_fpn_160_2x_ms_coco.py /path/to/your/save_dir/cascade_opa_522.pth 8 --eval bbox
## Acknowledgement
This repo is developed based on [mmdeteion](https://github.com/open-mmlab/mmdetection) and [SEPC](https://github.com/jshilong/SEPC). Please check mmdetection for more details and features.