KAN See In the Dark [paper]

Get Started

Dependencies and Installation

• Python 3.8
• Pytorch 1.11

1. Create Conda Environment

conda create --name KSID python=3.8
conda activate KSID

2. Install PyTorch

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113

3. Clone Repo

git clone https://github.com/AXNing/KSID.git

4. Install Dependencies

cd KSID
pip install -r requirements.txt

Data Preparation

Download the raw training and evaluation datasets

[Google Drive]

├── dataset
    ├── LOLv1
        ├── our485
            ├──low
            ├──high
	├── eval15
            ├──low
            ├──high
├── dataset
   ├── LOLv2
       ├── Real_captured
           ├── Train
           ├── Test

Testing

1. Our pre-trained weights can be downloaded by clicking here, and put it in the following folder:

├── checkpoints
    ├── lolv1_gen.pth
    ├── lolv2_real_gen.pth
    ├── lsrw_gen.pth

2. Modifying the pre-trained weight path in the json file.

    "path": {
        "log": "logs",
        "tb_logger": "tb_logger",
        "results": "results",
        "checkpoint": "checkpoint", 
        "resume_state": ""
    },

# LOLv1
python test.py --dataset ./config/lolv1.yml --config ./config/lolv1_test.json

# LOLv2-real
python test.py --dataset ./config/lolv2_real.yml --config ./config/lolv2_real_test.json

3. Evaluation metrics are referenced here.

Training

train uncertainty model

python train.py -uncertainty --config config/llie_train_u.json --dataset config/lolv1.yml

train the second stage

Modifying the uncertainty model weight path in the json file.

    "path": {
        "log": "logs",
        "tb_logger": "tb_logger",
        "results": "results",
        "checkpoint": "checkpoint", 
        "resume_state": ""
    },

python train.py --config config/lolv1_train.json --dataset config/lolv1.yml

To Do List

• Release the testing code for KSID.
• Upload the pretrained checkpoints.
• Release the training code for KSID.

Acknowledgement

Our code is built upon GSAD. Thanks to the contributors for their great work.