This project was made for hacathon conducted by IITG.AI club, IIT Guwahati.
I achieved rank 2 in this hackathon.
This AI works on basis of Autoencoders. It works using bottleneck architecture. First the Convolutional Layers works as encoders and Transposed Convolutional Layers decodes the image.
It converts gray scale image to LAB color space.
It also uses skip connections from encoder layers to decoder layers to avoid the problem of vanishing gradients and so that gradient can backpropogate faster.
totals params= 100k Loss function: MSE Optimiser: RMSprop output activation: Sigmoid
| Input | output |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |