Welcome! This is the official implementation of our paper: Multi-scale Synergism Ensemble Progressive and Contrastive Investigation for Image Restoration, published in IEEE Transactions on Instrumentation and Measurement.
Authors: Zhiying Jiang†, Shuzhou Yang†, Jinyuan Liu, Xin Fan, Risheng Liu* (†equal contribution, *corresponding author)
- Linux or macOS
- Python 3.7
- NVIDIA GPU + CUDA CuDNN
Type the command:
conda create -n SEPC python=3.7
conda activate SEPC
pip install -r requirements.txt
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.3 -c pytorch
You need create a directory ./logs/[YOUR-MODEL]
(e.g., ./logs/SEPC_derainL
).
Download the pre-trained model and put it into ./logs/[YOUR-MODEL]
.
Here we release the pre-trained model trained on Rain100L and Rain100H:
- You need create a directory
./testData
and put the degraded images to it. - Test the model with the pre-trained weights:
CUDA_VISIBLE_DEVICES=0 python test.py
- The test results will be saved to a directory here:
./results
.
- You need create a directory
./trainData
and put the degraded training data to it. - Train a model:
CUDA_VISIBLE_DEVICES=0 python train.py
- Loss curve and checkpoint can be found in the directory
./log
.
If you find this code useful for your research, please use the following BibTeX entry.
@ARTICLE{10363208,
author={Jiang, Zhiying and Yang, Shuzhou and Liu, Jinyuan and Fan, Xin and Liu, Risheng},
journal={IEEE Transactions on Instrumentation and Measurement},
title={Multiscale Synergism Ensemble Progressive and Contrastive Investigation for Image Restoration},
year={2024},
volume={73},
number={},
pages={1-14},
doi={10.1109/TIM.2023.3343823}}