I want to inference caffe model with INT8 precision, what should I do

Autonomous Machines Jetson & Embedded Systems Jetson Xavier NX
1

HI,
I want to inference caffe model with INT8 precision, but the Demo provided using onnx model. So I want to know wherever have a demo to study.

I also can’t understand why it need to set dynamic range for the tensor and set the output type of this layer

Thanks for your time.

2

is INT8 just supported by onnx model?

4

Hi,

No.
These are independent function. All model format can be inferenced with INT8.
You can simply use our trtexec binary to test different model format with different precision mode.

$ /usr/src/tensorrt/bin/trtexec --help

Thanks.