- A contest about image denoising
- For details, see [知乎]中兴捧月-图像去噪
├── checkpoints
├── data
│ ├── train
│ │ ├── ground_truth
│ │ └── noisy
│ └── valid
│ ├── ground_truth
│ └── noisy
├── demo_code
│ ├── dataloader
│ │ ├── data_augment.py
│ │ ├── data_loader.py
│ │ └── data_process.py
│ ├── net
│ │ ├── unet.py
│ │ ├── unet++.py
│ │ └── unet_acnet.py
│ ├── losses
│ │ └── losses.py
│ ├── main.py
│ ├── test.py
│ ├── test_metrics.py
│ ├── test_unet.py
│ ├── train_unet.py
│ ├── utils.py
│ └── valid.py
└── requirements.txt
- ubuntu16.04 + pytorch1.9.0+ cuda10.2 + python3.8
- You might need 18g memory if you're training with whole image
- You can try splitting image training, the code has been open source
- Run
conda create -n zte_contest python=3.8
to create virtual environment - Run
pip install -r requirement.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
to install required modules.
- In the
demo_code
folder, run
python main.py
-
Place trained model at the
checkpoints/UNet/
-
In the
demo_code
folder, run
python test.py