/SketchX

A GAN that converts sketches into potraits following pix2pix approach

Primary LanguageJupyter Notebook

SketchX: Sketch to image transformation using generative adverserial networks

Abstract

This project aims to develop a model that can convert sketches into portraits of Indian people using deep image-to-image translation techniques. Converting a sketch into a portrait photo is done by implementing a pix2pix approach to conditional GANs, with the generator following a U-net architecture and the discriminator a PatchGAN. The system is trained on a dataset containing faces of Indian people, ensuring that the generated image will also be of Indian origin. The results are evaluated based on the Fréchet Inception Distance between the real and generated image.

Results of the model

image

FID Score

image