README.md

Squeeze-and-Excitation Networks Reference Apps using TensorRT 5 and DeepStream SDK 3.0

Squeeze-and-Excitation Networks

1. Introductions and implementations from the author: paper and Repository

2. The TensorFlow re-implementation from which the checkpoints and TensorBoard are extracted for TensorRT inference: Repository

Getting Started

This repository provides the SENet reference applications using TensorRT 5 and DeepStream SDK 3.0.

There are two sample applications under senet/apps : trt-sent and deeptream-resnet-senet.

1. trt-senet-app is the stand alone application using TensorRT and SENet.

   Further details are provided in here
   To use only the stand alone trt-senet-app, Deepstream installation can be skipped.
   

2. deepstream-senet-app is the reference application using Deepstream and SENet.

   SENet is used as a secondary inference in this example.
   In order to run this app, you must run trt-senet-app to obtain TensorRT engine for SENet.
   In addition, you must install Deepstream and other pre-requisites.
   Further details are provided in here