The paper[Link] is accepted by ACMMM2020(oral). This is pytorch implementation for our paper. The code is based on the Pix2Pix.
- python 3.6
- pytorch 1.1.0
- opencv-python
- visdom
You could directly find our pretrained models in the directory.
To train the model, please run
python train.py --dataroot [your dataset path] --name rpdnet_lstm_100l --model rpdnet --gpu_ids 3
--batch_size 18 --output_nc 3 --input_nc 3 --niter 100 --niter_decay 0 --save_epoch_freq 10
--display_freq 40 --dataset_mode [rain100l|rain100h|rain1400] --display_freq 20 --lr 2e-4
--color_space 'rgb' --no_html --display_env lstm
To generate images, please run
python eval.py --dataset_path [your dataset path] --model_path [your model path]
--save_path [your save path] --gpu_id 0
The evaluation metrics are provided by Ren. The performances on the four datasets are listed below:
| Dataset | PSNR | SSIM |
|---|---|---|
| Rain100L | 38.80 | 0.984 |
| Rain100H | 30.33 | 0.909 |
| Rain1400 | 32.80 | 0.946 |
| Rain12 | 37.42 | 0.967 |