MachineLP / tensorrt-demo

TensorRT demo

TensorRT Demo

The demo shows how to build, train and test a conv-net using TensorFlow and then how to port it to TensorRT for fast inference.

Tensorflow part

Requirements: TensorFlow 1.4 (https://www.tensorflow.org/)

Tensorflow_train_model.ipynb -> Build and train the network using Tensorflow

Tensorflow_freeze_graph.ipynb -> Prepare the network for the inference procedure (mostly required for the further porting to TensorRT)

Tensorflow_inference.ipynb -> Inference by means of TensorFlow

TensorRT part

Requirements: TensorRT 3.0 (https://developer.nvidia.com/tensorrt): TensorRT and uff python libs. Python installation packages could be found in the TensorRT archive.

TensorRT_build_engine.ipynb -> Optimize frozen TF graph and prepare inference engine with TensorRT

TensorRT_inference.ipynb -> Inference by means of TensorRT