Skip to content

Nearly complete submissions for Super Resolution Convolutional Neural Network (SRCNN) algorithm.

Notifications You must be signed in to change notification settings

SirRacha/SRCNN_Satellite

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

These directories contain different approaches to working with SRCNN.

They include nearly complete submissions for Super Resolution Convolutional Neural Network (SRCNN) algorithm.

SRCNN-Tensorflow is our main focus for completing this application.


SRCNN-COLAB showcases how to go about setting up Tensorflow models in the Google Colab environment.


SRCNN-Keras is a nonworking proof of concept.


Model Architecture

Super Resolution CNN (SRCNN)

The model above is take from user titu1994 who I cite as a source for this project.

I have some differences from the original paper, and from other scientists who have worked on this topic:
[1] Used the Adam optimizer instead of Stochastic Gradient Descent.
[2] Stride is set to 21 instead 14.

My models underperform compared to the results posted in the paper. I was unable to replicate the GPU power and time for running to get my PSNR as high, ie. 32dbs.

I have optimized this model for use on different imagery instead of the usual 91 images from ImageNET.