Skip to content

Earthnook/deepstream_reference_apps

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

51 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Reference Apps for Video Analytics using TensorRT 5 and DeepStream SDK 3.0

DS3 Workflow

Installing Pre-requisites:

Install GStreamer pre-requisites using:
$ sudo apt-get install libgstreamer-plugins-base1.0-dev libgstreamer1.0-dev

Install Google flags using:
$ sudo apt-get install libgflags-dev

To use just the stand alone trt-yolo-app, Deepstream Installation can be skipped. Refer to the table below for information regarding installation dependencies. See Note for additional installation caveats.

Platform Applications Deepstream Version Dependancies Repo Release Tag
dGPU All apps and plugins DS 3.0 DS 3.0, Cuda 10, TensorRT 5 TOT
dGPU All apps and plugins DS 2.0 DS 2.0, Cuda 9.2, TensorRT 4 DS2 Release
dGPU trt-yolo-app Not required Cuda 10, TensorRT 5 TOT
jetson-TX1/TX2 nvgstiva-app, yoloplugin and trt-yolo-app DS 1.5 DS 1.5, Jetpack 3.3 (Cuda 9.0, TensorRT 4) DS2 Release
jetson-TX1/TX2 trt-yolo-app Not required Jetpack 3.3 (Cuda 9.0, TensorRT 4) DS2 Release
jetson-Xavier yoloplugin DS 3.0 Jetpack 4.1 (Cuda 10.0, TensorRT 5) TOT
jetson-Xavier trt-yolo-app DS 3.0 Jetpack 4.1 (Cuda 10.0, TensorRT 5) TOT

Tegra users working with Deepstream 1.5 and Jetpack 3.3 will have to regenerate the .cache files to use the standard caffe models available in the SDK. This can be done by deleting all the .cache files in /home/nvidia/Model directory and all its subdirectories and then running the nvgstiva-app using the default config file.

Setup

Update all the parameters in Makefile.config file present in the root directory

Building NvYolo Plugin

  1. Go to the data/yolo directory and add your yolo .cfg and .weights file.

    For yolo v2,
    $ wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov2.cfg
    $ wget https://pjreddie.com/media/files/yolov2.weights

    For yolo v2 tiny,
    $ wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov2-tiny.cfg
    $ wget https://pjreddie.com/media/files/yolov2-tiny.weights

    For yolo v3,
    $ wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg
    $ wget https://pjreddie.com/media/files/yolov3.weights

    For yolo v3 tiny,
    $ wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3-tiny.cfg
    $ wget https://pjreddie.com/media/files/yolov3-tiny.weights

  2. Add absolute paths of images to be used for calibration in the calibration_images.txt file within the data/yolo directory.

  3. For dGPU's,
    $cd sources/plugins/gst-yoloplugin-tesla
    make && sudo make install

  4. For jetson,
    $cd sources/plugins/gst-yoloplugin-tegra
    make && sudo make install

Object Detection using DeepStream

sample output

There are multiple apps that can be used to perform object detection in deepstream.

deepstream-yolo-app

The deepStream-yolo-app located at sources/apps/deepstream_yolo is a sample app similar to the Test-1 & Test-2 apps available in the DeepStream SDK. Using the yolo app we build a sample gstreamer pipeline using various components like H264 parser, Decoder, Video Converter, OSD and Yolo plugin to run inference on an elementary h264 video stream.

$ cd sources/apps/deepstream-yolo
$ make && sudo make install
$ cd ../../../
$ deepstream-yolo-app <Platform-Telsa/Tegra> <H264 filename> <yolo-plugin config file>

Refer to sample config files yolov2.txt, yolov2-tiny.txt, yolov3.txt and yolov3-tiny.txt in config/ directory.

trt-yolo-app

The trt-yolo-app located at sources/apps/trt-yolo is a sample standalone app, which can be used to run inference on test images. This app does not have any deepstream dependencies and can be built independently.

$ cd sources/apps/trt-yolo
$ make && sudo make install
$ cd ../../../
$ trt-yolo-app --flagfile=/path/to/config-file.txt

Refer to sample config files yolov2.txt, yolov2-tiny.txt, yolov3.txt and yolov3-tiny.txt in config/ directory.
Additionally run $ trt-yolo-app --help for a complete list of config parameters.

Note

  1. If you want to use the nvyolo plugin with the deepstream-app, you will need to modify the deepstream-apps' source code to read nvyolo plugins' properties and add it to the pipeline. You can refer the ds-example plugin in deepstream_config_file_parser.c and make equivalent changes required for the nvyolo plugin. Specifically refer to parse_dsexample function and its usage in deepstream_app_config_parser.c

About

Samples for TensorRT/Deepstream for Tesla & Jetson

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C++ 89.8%
  • Makefile 7.5%
  • Cuda 2.7%