Neural-Hacker/hand_gesture_yolo

Hand Gesture YOLOv8 (ONNX INT8)

Overview

This repository contains a YOLOv8 nano model trained for real-time hand gesture detection and optimized for edge and mobile devices.

The model is exported to ONNX format and quantized to INT8 for efficient on-device inference.

Model Details

• Architecture: YOLOv8n (Nano)
• Task: Hand gesture detection
• Format: ONNX
• Precision: INT8 (W8A8)
• Input size: 640 × 640
• Classes:
  • hand-gestures
  • left
  • right
  • stop
  • mvefrd

Included Assets

• hand_gesture_yolo_onnx_w8a8.zip
  • model.onnx
  • tool_versions.yaml
  • data.yaml

Each ZIP follows the Qualcomm AI Hub Community contribution guidelines.

Training Summary

• Dataset size: 330 images
• Training platform: Kaggle (NVIDIA Tesla T4)
• Framework: Ultralytics YOLOv8
• Quantization: ONNX INT8

Intended Use

• Edge AI applications
• Gesture-based interfaces
• Touchless controls
• Academic and research purposes

Limitations

• The model is trained on a small dataset and may not generalize to all lighting conditions or camera angles.
• Performance may vary depending on device and runtime.

License

This model is released under the CC BY-NC 4.0 license.

Disclaimer

Commercial use is not permitted.