deep-learning-in-scale-workshop

Goal

• Discover approach to inject Computer Vision algorithms in Video Streaming pipeline.

Real-world application

• Safe city program in Vinnytsia, Ukraine

Requirements

• Gstreamer (>=1.14)
• Tensorflow (>=1.10)
• OpenCV (>= 3.4)
• Ubuntu (18.04)
• Python (>=3.5)

Installation

Docker

docker pull jackersson/ubuntu-gstreamer
    
# Enable UI support in Docker Container 
xhost +local:

sudo docker run --name workshop -it -e DISPLAY=$DISPLAY -v /tmp/.X11-unix/:/tmp/.X11-unix jackersson/ubuntu-gstreamer

sudo docker exec -e DISPLAY=$DISPLAY -it workshop /bin/bash

cd /home/workshop/deep-learning-in-scale-workshop
source venv/bin/activate 
export GST_PYTHON_PATH=../gst-python/plugin/

Git

git clone https://github.com/jackersson/deep-learning-in-scale-workshop.git
cd deep-learning-in-scale-workshop
make install
source venv/bin/activate

Guide

How to inject python plugin in Gstreamer Pipeline?

• code

  ./workshop/gst_python_plugin_template.sh
  

How to implement simple blur filter with python plugin in Gstreamer pipeline?

• code

  ./workshop/gst_python_blur_plugin.sh
  

How to implement python plugin with Tensorflow model in Gstreamer pipeline?

• code

  ./workshop/gst_tf_detection_to_console.sh
  

How to implement python plugin to draw on image buffer in Gstreamer pipeline?

• code

  ./workshop/gst_tf_detection_overlay.sh
  

How to process multiple video sources with Tensorflow Model?

• code

  ./workshop/gst_tf_detection_multi.sh
  

How to launch Gstreamer Pipeline with python plugin with Shared Tensorflow Model?

• code

  ./workshop/gst_video_surv.sh
  

How to implement batch forward pass for multiple video streams

• code

  ./workshop/gst_video_surv_batch.sh