CycleGAN for Person Face Sketches: Face-to-Sketch and Sketch-to-Face Translation

Overview

This project implements CycleGAN for image-to-image translation between real human faces and their corresponding sketches. The model is trained to:

• Convert a real face image into a sketch.
• Convert a sketch back into a real face image.

The CycleGAN model is trained end-to-end using the Person Face Sketches dataset and includes a Flask-based API for real-time image conversion.

---

Key Objectives

• Train CycleGAN to map sketches to real faces & vice versa.
• Handle memory limitations by implementing batch-wise data loading.
• Save & resume model training using checkpointing.
• Deploy the trained model via a simple Flask-based UI.
• Allow users to upload images (or capture from a camera) for conversion.

---

Dataset Details

• Dataset Source: Person Face Sketches (Kaggle)
• Total Images: ~23K (Paired real-face & sketch images)
• Training Set: 20.7K images per category (photos & sketches)
• Validation Set: 1K images per category
• Test Set: 679 images per category
• Format: Grayscale sketches paired with RGB photos

---

Repository Contents

• i201819_GenAI_B_A2_Q4.ipynb → Jupyter Notebook containing CycleGAN training & inference
• app.py → Flask-based API to allow users to upload an image and perform real-time face-to-sketch or sketch-to-face conversion using the trained CycleGAN model
• index.html → Web interface for user interaction, allowing users to upload images and view the generated outputs
• i201819_ImamaAmjad_Ass2.pdf → Detailed analysis, methodology, and results
• README.md → Project documentation (to be expanded)

For now, please refer to the i201819_ImamaAmjad_Ass2.pdf for dataset details, model training methodology, and evaluation metrics. The README will be expanded later with additional explanations and improvements.

---

Future Enhancements

✅ Expand the README with dataset details, preprocessing steps, and architectural explanations
✅ Include sample outputs & qualitative comparisons between real faces and generated sketches
✅ Add challenges faced and key learnings section
✅ Optimize model training with better hyperparameters & additional GAN tricks
✅ Enhance UI for a smoother user experience
✅ Deploy as a web app or API for real-time inference

---