A Bright Feature Selection Transformer for Single-Image Deraining

Training

• Please download the corresponding training datasets and and modify the path in train.yml.
• Different datasets may require different training hyper-parameters. Please train the model with the arguments mentioned in the paper.
• Before training the model, you need to open the visdom server in another command window to visualize the deraining results, please run:

python -m visdom.server -port=2333

• Follow the instructions below to begin training our model.

python train.py

Run the script then you can find the generated model weights in the folder 'logs'.

Testing

• Please use the corresponding testing datasets and and modify the path in test.yml.
• Follow the instructions below to begin testing our model.

python test.py

Run the script then you can find the single-image deraining results in the folder 'logs'.

Pre-trained Models

Our pre-trained models are available for download in Baidu Netdisk: Download Link
Extraction Code: 3q1f

Visual Deraining Results

Our deraining results are available for download in Baidu Netdisk: Download Link
Extraction Code: fjaq