Image style transfer is a technique that blends the artistic features of a style reference image with the content of a target image. This project aims to explore and implement image style transfer using Generative Adversarial Networks, comparing their performance with traditional Convolutional Neural Networks. We focused on CycleGAN and MSG-Net for our implementation, examining how these state-of-the-art methods enhance the quality and realism of style-transferred images.
Weights used or generated from this project are available at this Google Drive: https://drive.google.com/drive/folders/1SfgepgKdAJE5g8FA2I6HP3axY423K_Xn?usp=sharing
To run, download the weights and place them in the folder ./models/output/*
| Name | Student ID | |
|---|---|---|
| Elijah Maron | z5372352@ad.unsw.edu.au | z5372352 |
| Hari Birudavolu | z5419889@ad.unsw.edu.au | z5419889 |
| Michael Girikallo | z5416925@ad.unsw.edu.au | z5416925 |
| Tianshuo Xu | z5358205@ad.unsw.edu.au | z5358205 |
| Vincent Pham | z5363266@ad.unsw.edu.au | z5363266 |