image-learner
image-learner copied to clipboard
schenker
Training a neural network to map from x,y of an images pixels to r,g,b.
Learning to reconstruct an image from pixel coordinates
Train a neural network to map from x and y coordinate of each pixel to the pixels r,g,b values. The network has 2 inputs (x and y), several fully connected (highway) layers and three outputs (r, g and b).
See implementation in image-learner.ipynb
Examples
Input:
Sample Outputs:
after 12 epochs with a network of width 6 and depth 35
after 520 epochs with a network of width 150 and depth 5
Input:
Sample Outputs:
after 4 epochs with a network of width 6 and depth 25
after 14 epochs with a network of width 6 and depth 25
Input:
Sample Outputs:
after 2 epochs with a network of width 50 and depth 20
after 142 epochs with a network of width 50 and depth 20
Input:
Sample Outputs:
after 7 epochs with a network of width 6 and depth 25
)
after 45 epochs with a network of width 6 and depth 25
Input:
Sample Outputs:
after 8 epochs with a network of width 70 and depth 5
after 75 epochs with a network of width 70 and depth 5
schenker