This is a repository for the 10701 project "Blur Kernel Estimation and Tracking via GANs". The
git clone --recurse-submodules https://github.com/Harsharma2308/DeepDeblur.git
conda create --name py37 python=3.7
- NVIDIA GPU + CUDA CuDNN
- Pytorch
pip install -r requirements.txt
cd DeepDeblur/Blind-Image-Deconvolution-using-Deep-Generative-Priors/scripts
python svhn_train.py
python blur_train.py
python DCGAN_train.py
cd gan_train
Run `deblurring_SVHN_wo_init.py` for running the algorithm without the modified initialisation of latent space.
Run `deblurring_SVHN_with_init.py` for running the algorithm with initialisation of latent space.
pip install -r requirements.txt
cd gan_train/data
wget http://ufldl.stanford.edu/housenumbers/train_32x32.mat
cd Blind-Image-Deconvolution-using-Deep-Generative-Priors/blur_data_generation
matlab -nodisplay -nodesktop -r "run blur_data_generate.m"
REGULARIZORS = [0.01 , 0.01] alpha = 1.0 (for Algorithm 2) NOISE_STD = 0.01 STEPS = 6000