Jetson Nano convert tensorflow model to tensorrt

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
,
1

I want to convert my image segmentation model created in tensorflow on a windows machine to tensorrt so I can use it on my jetson nano.

I have tried to use the library:
from tensorflow.python.compiler.tensorrt import trt_convert as trt
on the nano but it’s not powerful enough because even if I leave it working for two days it’s not able to convert it

(to do this I have relied on the nvidia documentation)

On my windows machine I can’t convert it because the conversion library is not available on windows

So I am trying to install Ubuntu with Cuda, cudnn, Tensorrt and Tensorflow but I am having a lot of problems to install Tensorflow with Tensorrt support.

Is there any alternative method that I haven’t thought of?

3

Hi,

Is it possible to convert the model into ONNX on Windows?
Usually, this can be done via tf2onnx tool.

If yes, ONNX model can be run with TensorRT binary directly.

$ /usr/src/tensorrt/bin/trtexec --onnx=[filename]

Thanks.

4

I could try,

So if i convert it to onnx the model would run as a “native” TensorRT model?

Thanks

5

Hi,

Yes, after the conversion, the model can work with TensorRT directly.
No TensorFlow library is required.

Thanks.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.