This repository is for the work of SMCycleGAN, which is built based on CycleGAN and used for translating artistic portraits to photo-realistic visualizations. This model is developed for the course project in COMP4711: Deep Learning for Computer Vision, HKUST, by YIN Zhuohao. This model is compared with several state of the art models such as CycleGAN, Art2Real, and has produced much more compelling results, as shown below.
To reproduce the results and download pre-trained models, you may check out this folder on Google Drive. It contains datasets, model checkpoints, and generated examples.
You may check out the project final report here on arXiv.