how to increase accuracy though esp-detection?

[xpdrycrytest]

Checklist

• Checked the issue tracker for similar issues to ensure this is not a duplicate.
• Described the feature in detail and justified the reason for the request.
• Provided specific use cases and examples.

Feature description

how to increase accuracy though esp-detection? quantize_onnx_model.py will decrease accuracy!!

Use cases

how to increase accuracy though esp-detection? quantize_onnx_model.py will decrease accuracy!!

Alternatives

No response

Additional context

No response

[Ginosko-mia]

Hi, @xpdrycrytest .
Some accuracy drop is expected when quantizing a model from float to int. To boost quantized model accuracy, you can consider several strategies:
1. Use int16 for layers that exhibit high layer-wise quantization error. Please refer to here.
2. Apply layerwise equalization or other optimization passes to improve activation and weight distribution, esp-detection applied LayerwiseEqualization .
3. Train the model with Quantization-Aware Training (QAT) for better accuracy retention.