What to know before use: 1、I have tested this project, which can be compiled and tested correctly. The problems in the above process have nothing to do with my code. 2、The evaluate result is coco. If you use DOTA, you need to convert the format yourself. 3、If training is unusual, pay attention to the relationship between your batch size and learning rate. 4、Many users can use this code normally. I have changed my profession and no longer maintain this project. Please do not open strange issue again.
Official implementation of Learning Point-guided Localization for Detection in Remote Sensing Images
In this repository, we release the OPLD code in Pytorch.
- OPLD architecture:
- OPLD output on DOTA:
- 4 x TITAN X GPU
- pytorch1.1
- python3.6.8
Install OPLD following INSTALL.md.
| Model | LR | mAP50 | FPS | DOWNLOAD |
|---|---|---|---|---|
| R-101-FPN_MS | 1x | 76.43 | 5.2 | GoogleDrive, BaiduNetDisk (4pt9) |
To train a model with 4 GPUs run:
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 tools/train_net.py --cfg cfgs/DOTA/e2e_OPLD_R-50-FPN_1x.yaml
python tools/test_net.py --cfg ckpts/DOTA/e2e_OPLD_R-50-FPN_1x/e2e_OPLD_R-50-FPN_1x.yaml --gpu_id 0,1,2,3
python tools/test_net.py --cfg ckpts/DOTA/e2e_OPLD_R-50-FPN_1x/e2e_OPLD_R-50-FPN_1x.yaml --gpu_id 0
If you use DOTA dataset and find this repo useful, please consider cite.
@inproceedings{xia2018dota,
title={DOTA: A large-scale dataset for object detection in aerial images},
author={Xia, Gui-Song and Bai, Xiang and Ding, Jian and Zhu, Zhen and Belongie, Serge and Luo, Jiebo and Datcu, Mihai and Pelillo, Marcello and Zhang, Liangpei},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={3974--3983},
year={2018}
}
@ARTICLE{9252176,
title={Learning Point-guided Localization for Detection in Remote Sensing Images},
author={Q. {Song} and F. {Yang} and L. {Yang} and C. {Liu} and M. {Hu} and L. {Xia}},
journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
year={2020},
}
OPLD is released under the MIT license.