- The data set for this task is CIFAR-10
- Using a tensorflow backend
- From 6000 images, we chose 10% of the pixels where each pixel is an RGB vector with three elements.
- We use clusetring to obatin the outputs of the network by converting the colored images to k-colored images
- The input of the network is created by converting the original image to grayscale
- For the CNN, we use 2 convolution layers, 2 MLP layers with 5 * 5 filters and a softmax layer, and one max pooling layer
nikhitmago/deep-cnn-for-image-colorization
Colorize an image from grayscale using Convolutional Neural Networks
Jupyter NotebookMIT