Image Super-Resolution Using Deep Convolutional Networks

Tensorflow implementation of SRCNN.

Prerequisites

• Python 3
• Tensorflow
• Numpy
• Scipy
• Opencv 3
• h5py

Usage

To train, uncomment the scripts in the bottom in net.py. Then type python net.py
To test, set proper img_path, save_path and upscaling factor (multiplier) in the use_SRCNN.py. Then type python use_SRCNN.py

Results

The following results are based on 45 hours of training on my i7 CPU.

Bicubic interpolation:

SRCNN:

Bicubic interpolation:

SRCNN:

Bicubic interpolation:

SRCNN:

We can also feed any image to this model to get an upscaled version with interpolated details:
Original image:

SRCNN:

Reference:

• Dong, C., Loy, C.C., He, K., Tang, X.: Learning a Deep Convolutional Network for Image Super-Resolution.
  
• tegg89/SRCNN-Tensorflow
  • • I have followed the loading and storing of h5 format files of this repository.