DPSHS

This is a pytorch implementation of our research. Please refer to our paper for more details: Deep Plug-and-play Nighttime Non-blind Deblurring with Saturated Pixel Handling Schemes

Dependencies

• python 3.8.12 (tested with anaconda3)
• PyTorch 1.10.0 (CUDA=11.3)

Install Packages

pip3 install -r requirement.txt -f https://download.pytorch.org/whl/cu113/torch_stable.html

Download the DRUNet pre-trained model from DPIR

Please put the drunet_color.pth below model/pre-trained.

Benchmark Datasets

• The Saturated Dataset from Hu et al.
• Low-illumination Dataset from Pan et al.
• Night Dataset from Chen et al.
• Download all datasets and then put them below the Benchmark/ folder.

Reproduce the Saturated Dataset

python3 test_deblur_benchmark.py -dataset Hu -log Log/Compare_Saturated/

Reproduce the Low-illumination Dataset

python3 test_deblur_benchmark.py -dataset Pan -log Log/Compare_Low/

Reproduce the Night Dataset

python3 test_deblur_benchmark.py -dataset Chen -log Log/Compare_Night/

Reproduce the Real World Blurry Images

python3 test_deblur_real_world.py

Note

• This implementation is based on the implementation of Pytorch-template and DPIR.