An unofficial implementation of CBDNet by PyTorch.
Download the dataset and pre-trained model: [OneDrive] [Baidu Pan (8ko0)] [Mega]
Extract the files to dataset folder and checkpoint folder as follow:
Train the model on synthetic noisy images:
python train_syn.py
Train the model on real noisy images:
python train_real.py
Train the model on synthetic noisy images and real noisy images:
python train_all.py
In order to reduce the time to read the images, it will save all the images in memory which requires large memory.
Test the trained model on DND dataset:
python test.py
Optional:
--ckpt {all,real,synthetic} checkpoint type
--cpu [CPU] Use CPU
Example:
python test.py --ckpt synthetic --cpu
Given a clean image x, the realistic noise model can be represented as:
Where y is the noisy image, f(.) is the CRF function and the irradiance ,
M(.) represents the function that convert sRGB image to Bayer image and DM(.) represents the demosaicing function.
If considering denosing on compressed images,
