rohit-pardasani / DeblurCNN

Convolutional Neural Network to deblur the images. A 20 layer DnCNN trained to remove blur introduced by averaging filter.

DeblurCNN

Convolutional Neural Network to deblur the images. A 20 layer DnCNN trained to remove blur introduced by averaging filter.

Architecture of DnCNN used

Convolution Layer - 1 (3 x 3 kernel) (Input Channels = 3) (Output Channels = 64) Activation (ReLU)

Convolution Layer - 2 (3 x 3 kernel) (Input Channels = 64) (Output Channels = 64) Batch Normalization Activation (ReLU)

Convolution Layer - 3 (3 x 3 kernel) (Input Channels = 64) (Output Channels = 64) Batch Normalization Activation (ReLU)

Convolution Layer - 4 (3 x 3 kernel) (Input Channels = 64) (Output Channels = 64) Batch Normalization Activation (ReLU)

...

...

...

...

...

...

Convolution Layer - 20 (3 x 3 kernel) (Input Channels = 64) (Output Channels = 3)

Some Results

Original Image	Blurred Image (Input to DeblurCNN)	Deblurred Image (Output of DeblurCNN)