Single Image Super Resolution

Project Aim

Implement image super resolution using various Neural Networks ie our network takes in a low resolution image and converts it to a high resolution image.

Source Code

All those interested can view the source code here

Members

• Prateek
• Sushruth
• Shreyas
• Rohith

Work Done

Initially we implemented a very basic CNN based on this research paper. We experimented on various hyperparameters like filter size,number of layers,Network depth,different datasets etc. Later we tried a different architecture based on this paper and similar tests for hyperparameters were conducted The key differences between the two networks were:

• Significantly more depth
• Training is on residual images ie difference between high and low res images

Results

Future Work

• Experiment with ResNet,InceptionNet or GANs for the same problem
• Build an applet for the same where users can input a low res image and a high res image will be outputted

Demo Video