This repository contains implementation of the CVPR - 2017 Paper "Image-to-Image Translation with Conditional Adversarial Networks".
The code is written in Python 3.7 and trained on Intel(R) Xeon(R) CPU @ 2.20GHz with Nvidia Tesla T4 15GB GPU
Part of Winter of GANs
The architecture is of U-Net kind, which takes input as a sketch images of size (256 X 256 X 3) and outputs a coloured image of size (256 X 256 X 3). Encoder layer consists of 8 layers which convert images into latent space of size (1 X 1 X 512). Decoder contains 7 layer which upsamples the image. L1 loss is used to the generator along with the cGan loss
It takes sketch as well as coloured images a input and stack one on another. It return the probability that given the sketch, does the coloured image belongs to it or not.
click here for getting the model weights. The model has been trained for 280 epochs, and the checkpoints can be found here if someone wishes to continue the training.
Link to the code in colab can be found here. Make a copy of the colab notebook in your drive and tinker around!!
Link to data used: https://www.kaggle.com/ktaebum/anime-sketch-colorization-pair
Some of the results
Made with ❤ by Bhuvan Aggarwal