DLDiff: Image Detail-guided Latent Diffusion Model for Low-Light Image Enhancement

🧊 Dataset

You can refer to the following links to download the LOLv1 as training data set.

Data Loader

If you want to change it, feel free to modify the /ldm/ldm/data/PIL_data.py to change the data loading format. During training, you need to replace the path of the training data set in the content of /DLDiff-main/ldm/config/low2light.yaml

🛠️ Environment

If you already have the ldm environment, please skip it

A suitable conda environment named ldm can be created and activated with:

conda env create -f environment.yaml
conda activate low2high

🌟 Pretrained Model

You can refer to the following links to download the sampling_model and the trainning_model available at Baidu Netdisk(rmk9) or Google Drive. Among them, the sampling_mode is used to predict results, and trainning_model is a pre-trained model used for model training. The pretrained model should be saved in the ./checkpoints/

🖥️ Inference

Prepare Testing Data:

you can download the LSRW(code: wmrr) and the LOLv2-real

Testing

Run the follwing codes:

bash predict.sh

The testing results will be saved in the ./results folder. The code includes modules for measuring PSNR, SSIM, FID, LPIPS, and time indicators. For FID measurement, ensure to download the ViT-B-32.pt model to the ./clip_model folder.

🧑‍💻 Train

Training with a 3090 GPU

Run the follwing codes:

bash train.sh

You can modify the paths of the config and checkpoints in the train.sh script. Example usage:

CUDA_VISIBLE_DEVICES=2 python main.py --base ldm/config/low2light.yaml --resume /DLDiff-main/checkpoints/train.ckpt --no_test False -t --gpus 0,