Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

End-to-End Conditional GAN-based Architectures for Image Colourisation

Aug 26, 2019
Marc G贸rriz, Marta Mrak, Alan F. Smeaton, Noel E. O'Connor

Share this with someone who'll enjoy it:

In this work recent advances in conditional adversarial networks are investigated to develop an end-to-end architecture based on Convolutional Neural Networks (CNNs) to directly map realistic colours to an input greyscale image. Observing that existing colourisation methods sometimes exhibit a lack of colourfulness, this paper proposes a method to improve colourisation results. In particular, the method uses Generative Adversarial Neural Networks (GANs) and focuses on improvement of training stability to enable better generalisation in large multi-class image datasets. Additionally, the integration of instance and batch normalisation layers in both generator and discriminator is introduced to the popular U-Net architecture, boosting the network capabilities to generalise the style changes of the content. The method has been tested using the ILSVRC 2012 dataset, achieving improved automatic colourisation results compared to other methods based on GANs.

* IEEE 21st International Workshop on Multimedia Signal Processing, 27-29 Sept 2019, Kuala Lumpur, Malaysia 

   Access Paper Source

Share this with someone who'll enjoy it: