/deep-cnn-for-image-colorization

Colorize an image from grayscale using Convolutional Neural Networks

Primary LanguageJupyter NotebookMIT LicenseMIT

Deep CNN for Image Colorization

  • 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