This is a tensorflow-2 implementation of the paper Beyond a Gaussian Denoiser: Residual Learning ofDeep CNN for Image Denoising.

BSDS200

To download and unpack run:

wget https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/BSDS300-images.tgz

tar -xvzf BSDS300-images.tgz

• tensorflow==2.3.1
• matplotlib
• numpy

Run train_denoise.py

After training by setting depth=5 for 25 epoch we get the below results. The original paper suggests a deeper architecture feel free to tweak the hyper params in the train_denoise.py file.