guanguanboy / SGF

Segmentation Guided Low-light Image Enhancement

News

• Nov 23, 2023: Codes, datasets, and pre-trained models will be released!

Installation

1. Clone our repository

Download our code.
cd SGF

2. Make conda environment

conda create -n pytorch111 python=3.7
conda activate pytorch111

3. Install dependencies

conda install pytorch=1.11 torchvision cudatoolkit=10.2 -c pytorch
pip install matplotlib scikit-learn scikit-image opencv-python yacs joblib natsort h5py tqdm
pip install einops gdown addict future lmdb numpy pyyaml requests scipy tb-nightly yapf lpips

4. Install basicsr

python setup_basicsr.py develop --no_cuda_ext

Prepare Dataset

Download the following datasets (including the edge maps generated by SAM):

LOL-v1 Baidu Disk (code: o42q), Google Drive

LOL-v2 Baidu Disk (code: g4c1), Google Drive

SID Baidu Disk (code: 9rlf), Google Drive

The code for generating edge maps using SAM are avaliable at (https://github.com/guanguanboy/SAM_Mask_Generation).

Training

Training instructions on different datasets are listed as follows.

LOL-v1

training code:

./train_sam.sh Enhancement/Options/Enhancement_SGF_Lolv1.yml

LOL-v2-real

training code:

 ./train_sam.sh Enhancement/Options/Enhancement_SGF_Lolv2_real.yml

LOL-v2-synthetic

training code:

./train_sam.sh Enhancement/Options/Enhancement_SGF_Lolv2_synthetic.yml

SID

training code:

./train_sam.sh Enhancement/Options/Enhancement_SGF_SID.yml

Evaluation

Fisrt download the pretained model from Google Drive or Baidu Disk（code: pacs）and put them in the root directory.

Evaluation instructions on different datasets are listed as follows.

LOL-v1

testing code:

python3 Enhancement/test_from_dataset.py --opt Enhancement/Options/Enhancement_SGF_Lolv1.yml --weights pretrained_models/LOLv1/net_g_latest.pth --dataset LOLv1_edge

LOL-v2-real

testing code:

python3 Enhancement/test_from_dataset.py --opt Enhancement/Options/Enhancement_SGF_Lolv2_real.yml --weights pretrained_models/LOLv2_real/net_g_latest.pth --dataset LOLv2

LOL-v2-synthetic

testing code:

python3 Enhancement/test_from_dataset.py --opt Enhancement/Options/Enhancement_SGF_Lolv2_synthetic.yml --weights pretrained_models/LOLv2_synthetic/net_g_latest.pth --dataset LOLv2_synthetic

SID

testing code:

python3 Enhancement/test_from_dataset.py --opt Enhancement/Options/Enhancement_SGF_SID.yml --weights pretrained_models/SID/net_g_latest.pth --dataset SID_SAM

Results

Experiments are performed for low-light image enhancement or four benchmark dataset.

Citation

If you use our code, please consider citing our paper:

Contact

Should you have any question, please contact [e-mail]

Acknowledgment: This code is based on the BasicSR toolbox.

Our Related Works

• Restormer: Efficient Transformer for High-Resolution Image Restoration, CVPR 2022. Paper | Code
• Learning Enriched Features for Real Image Restoration and Enhancement, ECCV 2020. Paper | Code