A Keras and Tensorflow implementation of super resolution using deep neural networks is proposed. The architecture is shown below.

• We divide the video temporal consecutive segments of 1 second duration
• We downsample the images of each segment and feed both downsampled and original image for training
• Each video segment has its own model (we called it as a micro-model because it is trained on a little data of 30 images)
• We overfit the model intentionally as there is no test phase

• Python3
• Tensorflow
• Keras
• Jupyter Notebook (optional)

Pranjal Sahu, Mallesham Dasari