This project is a demo application to show the capabilities of the embedded neural processing unit (NPU) of the i.MX 8MPlus in combination with the PHYTEC VM-016 camera.
The demo will recognize your face and extract embeddings via an artificial neural network (ANN). Your embeddings will be compared with precalculated embeddings from ~5k images of known Hollywood celebs. The ANN inference time lasts around a mean of 0.016s. The whole process takes a bit longer (~1s).
First you need to download and extract the required tensorflow lite model, previously created embeddings of the celebrities and their images.
wget https://download.phytec.de/Software/Linux/Applications/demo-celebrity-face-match-data-1.1.tar.gz
tar -xzf demo-celebrity-face-match-data-1.1.tar.gz
Second a PHYTEC VM-016 camera needs to be connected to the phyBOARD-Pollux MIPI-CSI2 CSI1 connector.
Alternatively a USB camera can be used. This requires following change in 'aidemo.py'
...
18 #CAMERA = 'VM-016'
19 CAMERA = 'USB'
...
Now the demo can be started with:
python3 aidemo.py
- python3
Following python modules:
- tflite_runtime (see https://www.tensorflow.org/lite/guide/python)
- opencv-contrib-python
- PyGObject
- pycairo
This project is licensed under the Apache License Version 2.0. See the file
LICENSE for detailed information.