Triangulation - Stereo neural inference demo

Because there are often application-specific host-side filtering to be done on the stereo neural inference results, and because these calculations are lightweight (i.e. could be done on an ESP32), we leave the triangulation itself to the host.

This 3D visualizer is for the facial landmarks demo, and uses OpenGL and OpenCV. Consider it a draft/reference at this point.

Demo uses 2-stage inferencing; 1st NN model is face-detection-retail-0004 and 2nd NN model is landmarks-regression-retail-0009.

Demo

Installation

sudo apt-get install python3-pygame
python3 -m pip install -r requirements.txt

Usage

Run the application

python3 main.py

You should see 5 windows appear:

• mono_left which will show camera output from left mono camera + face bounding box & facial landmarks
• mono_right which will show camera output from right mono camera + face bounding box & facial landmarks
• crop_left which will show 48x48 left cropped image that goes into the second NN + facial landmarsk that get outputed from the second NN
• crop_right which will show 48x48 right cropped image that goes into the second NN + facial landmarsk that get outputed from the second NN
• pygame window which will show the triangulation results