results are not as defined as examples
Opened this issue · 5 comments
Yes, results are depended on the input photos for the current model. What you mentioned are all possible improvements for the project, which were not my main focus at the time. It will be nice if you can come up with better models or more functionalities. Let me know or submit pull request when you do. Thanks!
@timxzz i would like to retrain the current model to improve it but not familiar with the current setup, where can I setup training test pair files as right now it is doing inference on the server which I am not familiar with how to setup for training
This is the complete code for the model, which includes both training and testing. And as mentioned in README, the code is based on TF version 1.3.0.
https://github.com/timxzz/GANs/blob/master/Face/Face_StarGAN_v5.ipynb
And this is for data preprocessing:
https://github.com/timxzz/GANs/tree/master/Face/Face_data
For detail of the algorithm and the training data, please refer for the project report:
http://timx.me/Facial_Expression_Modifier/pdf/Third_Year_Project_Report.pdf