PalmInference: Real-Time Palmistry Inference System

PalmInference is a real-time computer vision application that utilizes a custom-trained YOLOv8n model to detect, segment, and analyze key hand palm lines (Life, Heart, and Head) from live webcam feeds.

Designed for CPU-only environments, this project showcases the full Machine Learning lifecycle—from high-fidelity dataset curation to the deployment of optimized models using ONNX Runtime.

---

🚀 Key Features

• Real-Time Inference: High-speed detection pipeline optimized for consumer hardware using ONNX.
• Cultural & Behavioral Analysis: Translates visual features (line length, depth, curvature) into structured personality insights.
• Hardware Optimization: Leverages ONNX Runtime to bypass the need for heavy GPU compute, achieving significant speedups on standard CPUs.
• Quantization Strategy:
  • Implemented Static Quantization to minimize model footprint.
  • Actively managing a transition to Dynamic Quantization to recover precision (addressing a ~20% accuracy trade-off) while maintaining real-time latency.

---

🛠️ Technical Architecture

This project demonstrates advanced engineering trade-offs required for FinTech and high-performance AI solutions:

1. Data Curation: Custom dataset annotation for precise palm line segmentation.
2. Model Training: PyTorch-based training of YOLOv8 Nano architecture.
3. Deployment: Export to .onnx format for platform-agnostic inference.
4. Version Control: Rigorous Git workflow for feature branching and collaborative development.

---

📦 Installation

Prerequisites: Python 3.9+, Webcam.

# Clone the repository
git clone [https://github.com/Corneliox/PalmInference.git](https://github.com/Corneliox/PalmInference.git)

# Navigate to the directory
cd PalmInference

# Install dependencies (requires ultralytics, onnxruntime, opencv-python)
pip install -r requirements.txt

---

🖥️ Usage

To launch the inference engine with ONNX optimization enabled:

# Run on default webcam (Source 0)
python main.py --source 0 --model models/yolov8n-palm.onnx

To benchmark the quantized model performance:

python benchmark.py --quantized True

Quantization Still Under Developing

---

👥 Team & Collaboration

• Raka Surya Kusuma
  • Focus: Architecture, Model Training (PyTorch).
• Corneliox
  • Focus: Frontend Inference Logic, Behavioral Analysis Mapping, Visualization, ONNX Optimization, Quantization Strategy.

---

📄 License

Distributed under the MIT License. See LICENSE for more information.