TensorRT from inference graph

Hi all,

Device: Jetson TX2 - JP 3.3
TensorFlow v 1.9

I have an inference graph .pb of an image detector, I want to increase my FPS rate by optimizing the graph with TensorRT. I can’t find a clear explanation or a script to optimize my model from inference graph.

I know its not a hard question but please help if you can…

Hi,
In order to port .pb model file you need to convert it to UFF format and use UFF parser to generate the serialized engine files.
You can also use “trtexec” command line tool to benchmarking networks on random data and for generating serialized engines.

Please find the below reference/samples for working with TensorFlow:
(Assuming that the TRT version on your system is 5.1.x)

https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-515/tensorrt-developer-guide/index.html#import_tf_python

https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt-515/tensorrt-developer-guide/index.html#working_tf