Generative Adversarial Network (StyleGAN2-ADA)
Vu + Sloth (Vloth)
Start date
2021
Role
Lead Engineer
Project type
Generative Adversarial Network (StyleGAN2-ADA)
Project code
Report
Video Diary - Ideation Process (and a lot of rambling 😆)
Video Diary - Data Collation
Video Diary - Start to implement StyleGAN for better results
Video Diary - Successful implementation of StyleGAN
Built with
An Artificial Intelligence project using a method called Generative Adversarial Network (GAN), particularly StyleGAN2-ADA with a pretrained model on human faces, to train a custom model which can blend features of my face and sloth's face.
Please be aware the image resolution is not very good as it is, as far as I can remember, 80px-80px in resolution.
This is due to the constraint on hardware and graphics power to process and produce larger-sized images. It took me weeks just to get about 5-10 good results out of 1000 selfies and another 1000 images of sloths hanging on the trees, so imagine higher resolution 😁.
What I learned from this project is:
- What an abomination the results of this project have turned out to be 😆.
- But it is cool to see how with a pre-trained model integrated into your training, the training can become much easier, the result can become much more satisfying (turns out that I got the best grade for this out of all modules I took part in 🦥).
Gallery
