DemoFace

The collection of Jupyter notebooks and utilities for preparing image data and experimenting with vision+language models (inference and fine-tuning).

Quick start

1. Clone the repo:

   git clone https://github.com/Sufianlab/Demoface.git
   

2. Create & activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # macOS / Linux
   .\.venv\Scripts\activate    # Windows
   

3. Install basic dependencies:

   pip install jupyterlab torch torchvision transformers datasets pillow numpy opencv-python requests
   

4. Open notebooks:

   jupyter lab
   

Requirements

• Python 3.8+
• jupyter / jupyterlab
• torch, torchvision
• transformers, datasets
• pillow, numpy, opencv-python, requests
• Optional (for 8-bit / LoRA training): bitsandbytes, peft, accelerate

Notebooks (short)

• BLIP_2_inference.ipynb — BLIP-2 style inference examples
• BLIP_2_fine_tuneing.ipynb — fine-tuning with notes on LoRA / 8-bit workflows
• Paligemma_inference.ipynb — PaliGemma-style inference examples
• Paligemma_fine_tuneing.ipynb — fine-tuning pipeline for PaliGemma
• raw_image_downloader.ipynb — download raw demographic images with api
• face_preprocessing_through_cv2.ipynb — preprocessing with OpenCV / PIL

Dataset format

JSON entry format:

{
  "id": "123",
  "image": "path/to/image.png",   // or image bytes
  "Question": "Prompt text",
  "Answer": "Expected answer"
}

Ensure image paths are valid and update notebook variables that point to the dataset and image directories.

Dataset Access Form

You need to submit the Google Form below to get access to the dataset:

👉 Dataset Access Agreement Form

Disclaimer

The rest of the code and files are under construction and will be updated shortly.

Maintainer & Contact

Maintainer: Sufianlab

GitHub: https://github.com/Sufianlab