- Introduction
- Requirements
- Installation
- Pretrained Models and Logs
- Data Preparation
- Training
- Evaluation
- Citations
- License
- Acknowledgements
- Contact
This repository provides the code for the paper "FakeMix Augmentation Improves Transparent Object Detection".
- python3
- PyTorch=1.1.0
- torchvision
- Pillow
- numpy
- pyyaml
Please make sure that there is at least one gpu when compiling. Then run:
python3 setup.py develop
The pretrained models and logs can be fould here:
-
Baidu Drive with code: j3qp
- create dirs './datasets/Trans10K'
- download the data from Trans10K.
- put the train/validation/test data under './datasets/Trans10K'. The data Structure is shown below:
Trans10K/
├── test
│ ├── easy
│ └── hard
├── train
│ ├── images
│ └── masks
└── validation
├── easy
└── hard
CUDA_VISIBLE_DEVICES=0 python3 -u ./tools/test_demo.py --config-file configs/trans10K/trans10K.yaml DEMO_DIR ./demo/imgs
bash tools/dist_train.sh configs/trans10K/trans10K.yaml 8
CUDA_VISIBLE_DEVICES=0 python3 -u ./tools/test_translab.py --config-file configs/trans10K/trans10K.yaml
Please cite our paper if the project helps.
@article{cao2021fakemix,
title={FakeMix Augmentation Improves Transparent Object Detection},
author={Cao, Yang and Zhang, Zhengqiang and Xie, Enze and Hou, Qibin and Zhao, Kai and Luo, Xiangui and Tuo, Jian},
journal={arXiv preprint arXiv:2103.13279},
year={2021}
}
@misc{fanet,
author = {Cao, Yang and Zhang, Zhengqiang},
title = {fanet},
howpublished = {\url{https://github.com/yangcao1996/fanet}},
year ={2021}
}
For academic use, this project is licensed under the Apache 2.0 License
For commercial use, please contact the authors.
Our codes are mainly based on TransLab. Thanks to their wonderful works.
Any discussion is welcome. Please contact the authors:
Yang Cao: yangcao.cs@gmail.com
Zhengqiang Zhang: zhengqiang.zhang@hotmail.com