TensorRT vs TensorFlow-TRT

Hi

Where can I find a good documentation which discusses the differences between TensorRT and TensorFlow-TRT?

What will be faster TensorRT or TensorFlow-TRT (with a fully compatible graph) and Why?

Is there code available which shows the whole pipeline of functioning: From a TensorFlow model (Google’s Zoo) to the inference in the devboard (with TensorRT)

Thank you

Hi,

You can find some information here:
https://docs.nvidia.com/deeplearning/frameworks/tf-trt-user-guide/index.html

In general, pure TensorRT API can give you a better performance.

TF-TRT integrates TensorRT into the TensorFlow interface so you will need to create two implementation to enable fallback sometimes.
This mechanism will affect memory usage and performance but allowing user to use TensorRT easily.

Please check this GitHub for the TF-TRT samples:

Thanks.

1 Like