Enable Model Quantization for Performance Gains

[SurajSanap]

Files Affected:

• fire_detection.py
• gear_detection.py
• r_zone.py

📌 Current Model Info

• Models are running in FP32 precision (default).
• Framework: ultralytics.YOLO.
• No quantization or hardware-specific optimization applied.

🐞 Problem

• High memory consumption during inference.
• Slow performance on edge devices (Raspberry Pi, Jetson Nano, low-end CPU).
• Wasted compute power because lower precision can still maintain accuracy.

✅ Steps to Reproduce

1. Run any detection script on CPU-only hardware.
2. Observe memory usage and low FPS performance.

💡 Suggested Improvement

• Allow contributors to integrate quantized versions of YOLO:

  • FP16 (half precision)
  • INT8 (8-bit quantization)

• Add model export scripts to TensorRT / ONNX Runtime / TFLite for optimized inference.

• Provide a config flag like --quantized true to toggle quantized inference.

📊 Expected Outcome

• Up to 2–4× faster inference on supported devices.
• Reduced memory footprint.
• Better deployment support for edge hardware.

[Gupta-02]

Assign me this issue brother

[HimaleePatil]

Hello @SurajSanap ,
I'd want to contribute and work on this issue for GSSoC'25. Could you please assign this issue to me?