Google meet model run with openvino inference to optimize performance in intel x64 machine
- Process meet_camera get input from camera to do meeting stub standalone.(x86,x64...)
- Process meet_segment support virtual background for meet_camera.(only x64 support by openvino)
- These 2 process using share memory to communicate.
So this project solves the problem of architectural differences from x86 application and openvino toolkit support x64 app.
Awesome segment
demo.mp4
Low CPU on Lenovo Thinkpad i5-8250U
cpu_using.mp4
Just cpu processing, not using gpu
- Install openvino tool kit openvino_2021.2.185 and setup environment https://software.intel.com/content/www/us/en/develop/tools/openvino-toolkit.html
- Open meet_segment.sln visual studio Release x64
- Check general link include in project point to right place (openvino, opencv)
- Build meet_camera.exe & meet_segment.exe file
- Run deploy script to copy dll and resource file to exe folder
- Run meet_camera.exe in bin folder, meet_segment.exe auto start following
- https://github.com/PINTO0309/PINTO_model_zoo/tree/main/082_MediaPipe_Meet_Segmentation
- https://github.com/Volcomix/virtual-background
- https://software.intel.com/content/www/us/en/develop/tools/openvino-toolkit.html
Script to build IR from ONNX:
python mo.py --input_model meet3.onnx --input_shape [1,144,256,3] --scale 255 --data_type FP32