Skip to content

Latest commit

 

History

History
120 lines (96 loc) · 4.5 KB

README.md

File metadata and controls

120 lines (96 loc) · 4.5 KB

Tabla Drum Image Generator 🎵🎨

A Machine Learning Approach to Realistic Tabla Drum Image Generation

This project showcases AI/ML expertise through the development of a custom-trained image generation model focused on Indian tabla drums. By leveraging diffusion models and GAN architectures, the goal is to correct inaccuracies in existing AI-generated tabla drum images and produce high-quality, culturally accurate representations.

🚀 Project Goals

  • Curate & preprocess a high-quality dataset of tabla drum images.
  • Train a custom fine-tuned model using Stable Diffusion and GAN architectures.
  • Evaluate & optimize image fidelity using FID and perceptual loss metrics.
  • Deploy the model via a web-based API with a front-end for image generation.

🛠 Tools & Technologies

Machine Learning & AI

  • Frameworks: PyTorch, TensorFlow
  • Model Architectures: Stable Diffusion, DreamBooth, StyleGAN, GAN-based approaches
  • Data Augmentation: OpenCV, Albumentations

Data Collection & Processing

  • Web Scraping: Selenium, Playwright, Requests
  • Annotation: Label Studio
  • Storage: Cloud-based dataset hosting (AWS S3, Hugging Face Datasets)

Deployment & Serving

  • API Hosting: FastAPI, Flask
  • Web UI: Streamlit, Gradio, Hugging Face Spaces
  • Infrastructure: Docker, Kubernetes (planned for production-scale deployment)

Project Management & Version Control

  • GitHub Actions: CI/CD automation for training jobs & model updates
  • Experiment Tracking: Weights & Biases (planned integration)
  • Collaboration Tools: Notion, Trello

📂 Project Structure

tabla-image-gen-ai/
│── data_collection/        # Web scraping & dataset collection
│   ├── scraper/            # Selenium/Playwright-based image scraper
│   ├── images/             # Raw and processed dataset images
│   ├── metadata/           # Image metadata & annotations
│
│── data_processing/        # Dataset cleaning & augmentation pipeline
│   ├── preprocessing.py    # Image resizing, enhancement, noise reduction
│   ├── augmentation.py     # Data augmentation transformations
│
│── model_training/         # ML model training pipeline
│   ├── train.py            # Fine-tunes diffusion model or GAN on dataset
│   ├── evaluation.py       # Calculates FID, PSNR, SSIM
│
│── deployment/             # Web-based model serving
│   ├── api/                # FastAPI/Flask server for generating images
│   ├── ui/                 # Streamlit/Gradio front-end
│
│── notebooks/              # Jupyter notebooks for data exploration
│── app/requirements.txt    # Python dependencies
│── README.md               # This document

⚡ Setup Instructions

1️⃣ Clone this repository

git clone https://github.com/tyler-pritchard/tabla-image-gen-ai.git
cd tabla-image-gen-ai

2️⃣ Install dependencies

pip install -r app/requirements.txt

3️⃣ Run the web scraper (to collect tabla drum images)

python data_collection/scraper/tabla_image_scraper.py

4️⃣ Preprocess the images for model training

python data_processing/preprocessing.py

5️⃣ Train the AI model

python model_training/train.py

6️⃣ Deploy the image generation model

python deployment/api/main.py

📈 Current Progress

✅ Web scraping functional (collecting 100+ high-res tabla images)
✅ Dataset preprocessing implemented (image cleaning, augmentation)
🚧 Model fine-tuning in progress (Stable Diffusion adaptation)
🚧 Deployment infrastructure in planning

👨‍💻 Why This Project Matters

  1. Addresses a real-world AI failure: Existing generative AI models struggle with authentic tabla representations.
  2. Demonstrates ML proficiency: Covers data collection → model training → deployment.
  3. Production-ready deployment: Architected for real-world applications with Scalable APIs.
  4. Extensible for other domains: Can be adapted for any niche image dataset.

📩 Connect & Collaborate