Autoencoders—especially the convolutional version—can be used for harder tasks such as image colorization. In the following example, we feed the model with an input image without any colors, and the reconstructed version of this image will be colorized by the autoencoder model:
Figure 13: The CAE is trained to colorize the image
Figure 14: Colorization paper architecture
Now that our autoencoder is trained, we can use it to colorize pictures we have never seen before!
This kind of application can be used to color very old images that were taken in the early days of the camera.