seeker1943 / kerastotensorrt

This is a simple demonstration for running Keras model model on Tensorflow with TensorRT integration(TFTRT) or on TensorRT directly without invoking "freeze_graph.py".

Keras to TensorRT Examples

This is a simple demonstration for running Keras model model on Tensorflow with TensorRT integration(TFTRT) or on TensorRT directly without invoking "freeze_graph.py".

Note: Recommend that use NVIDIA Tensorflow docker image to run these examples. You can download the images from NVIDIA NGC.

Requirement

• Python (both 2 and 3 are ok)
• TensorRT (> 3.0)
• Tensorflow with TensorRT integration (> 1.7)
• Keras

if you want to run model on TensorRT directly, Pycuda is also needed:

• Pycuda (> 2017.1.1)

Examples

tftrt_example.py demonstrates how to run Keras model on TFTRT. This approach supports both NCHW and NHWC format because Tensorflow can handles format issue.

$ python tftrt_example.py

tftrt_resnet_example.py demonstrates how to run Keras Applications ResNet50 on TFTRT.

$ python tftrt_resnet_example.py

tftrt_multi_inputs_mutli_outputs_example.py demonstrates how to run a multi-input/output Keras model on TFTRT.

$ python tftrt_multi_inputs_mutli_outputs_example.py

trt_example.py demonstrates how to run Keras model on TensorRT which can achieve fastest speed. Because TensorRT didn't fully support NHWC yet, this approach only suits NCHW format.

$ python trt_example.py

Appendix

get_mnist_model.py can generate needed Keras models with two different input formats, one for NCHW foramt, another one for NHWC format.

Note: the needed models were already provided in repo.

$ python get_mnist_model.py -h # shows help message
$ python get_mnist_model.py -f 0 # generates model for NCHW format
$ python get_mnist_model.py -f 1 # generates model for NHWC format

Reference

• TensorRT Integration Speeds Up TensorFlow Inference
• NVIDIA Tensorflow Release Note