Tensorflow pointNet model conversion into tensorrt jetson nano

hello
Am trying to convert tensorflow model into tensorrt optimized model using the below code

converter = trt.TrtGraphConverterV2(
    input_saved_model_dir='saved_model',
    precision_mode='FP16',
    maximum_cached_engines=100,
    is_dynamic_op=True)

File ~/.local/lib/python3.8/site-packages/tensorflow/python/compiler/tensorrt/trt_convert.py:216, in _check_trt_version_compatibility()
211 if not _pywrap_py_utils.is_tensorrt_enabled():
212 logging.error(
213 "Tensorflow needs to be built with TensorRT support enabled to allow "
214 “TF-TRT to operate.”)
→ 216 raise RuntimeError(“Tensorflow has not been built with TensorRT support.”)
218 linked_version = _pywrap_py_utils.get_linked_tensorrt_version()
219 loaded_version = _pywrap_py_utils.get_loaded_tensorrt_version()

RuntimeError: Tensorflow has not been built with TensorRT support.

but I get an error of

Hi,

How do you install the TensorFlow package?
Are you using our prebuilt which has enabled the TF-TRT?

https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html

Thanks.

I used pip to install both tensorflow and tensorrt

Hi,

Please try the installation steps shared above.
The pip installation might link to a CPU-only package.

Thanks.

followed the procedure I still have the same error

hello AastaLLL

followed the same procedure but ends up with the same error

Hi,

Could you share the TensorFlow version (tf.version) with us?
Thanks.

hello AastaLLL

the tensorflow version is 2.12.0

Hi,

It looks like you still use the CPU TensorFlow since we don’t have such a version for Nano.

It seems that you are using python3.8. Is that correct?
We only provide the package for the default Python version.
For Nano, it is Python 3.6.

Thanks.

hi
I am making use of python 3.8.3,

whats the version compatible with it

best regards

Please could you tell me the best version of tensorflow and tensorrt that works for python 3.6

Am making use of jetpack version 32

Hi,

Please use the TensorRT comes with the same JetPack version.
For example, TensorRT 8.2.1 for JetPack 4.6.3.

For TensorFlow, please try our package that built with the same environment:
For example, below is the package for JetPack 4.6.3:
https://developer.download.nvidia.com/compute/redist/jp/v46/tensorflow/

Thanks.

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