728x90
mobileNet float32 → uInt8
- 예제
https://github.com/microsoft/onnxruntime-inference-examples
quantization → notebook → imagenet_v2 → mobilenet.ipynb
- memory size
# ONNX full precision model size (MB)
13.31911563873291
# ONNX quantized model size (MB)
3.4079103469848633
- runtime
# ONNX full precision model
tabby 0.70594245
Egyptian cat 0.14357993
tiger cat 0.12944907
lynx 0.0051510707
plastic bag 0.0037339667
# ONNX quantized model size
tabby 0.65022784
tiger cat 0.1949021
Egyptian cat 0.14421292
lynx 0.0047474923
tiger 0.0021263342
728x90
'Machine Learning > Edge Device' 카테고리의 다른 글
Install Apache TVM (0) | 2023.03.18 |
---|---|
ONNX & TensorRT & TVM (0) | 2023.03.18 |
경량화 기법 원리 ( 가지치기, 지식증류, 양자화 ) (0) | 2023.03.18 |
Docker에서 TensorRT 설치 (0) | 2022.09.02 |