zte_img_denoise

• A contest about image denoising
  
• For details, see [知乎]中兴捧月-图像去噪

Directory structure

├── 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

Running

Experimental environment

• 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

Start

• 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.

Train

• In the demo_code folder, run

  python main.py

Test

• Place trained model at the checkpoints/UNet/

• In the demo_code folder, run

  python test.py