hosted on hugging-face space now
Deployment of image to image translation model as a web-app.
Dataset - aligned CelebA
-
I used Contrastive Unpaired Translation (CUTGAN) for training the model.
-
Divided the data-set into two parts with eye-glasses and without eye-glasses based on provided attributes list.
-
For reducing complexity, portrait segmentation model is used to extract region of person and send it to the generative model as input.
| original | portrait | model output |
|---|---|---|
 |
 |
 |
CARN Model (Fast, Accurate, and Lightweight Super-Resolution with Cascading Residual Network) for up-scaling the output image.
- Python 3.5+
- PyTorch
- Tensorflow-lite (for portrait segmentation)
- Starlette
Run the web-app locally using command - python app/server.py serve
Contrastive Unpaired Translation (CUTGAN)
Portrait Segmentation models by Anil Sathyan
Waifu2x-PyTorch implementation
face restoration [latest addition for improving faces after generation]