This is an implementation of the visual wake words challenge winner at https://github.com/mit-han-lab/VWW.
The tflite file is a converted version of the trained float model and quantization is carried out in 8 bit activations and weights using sample data from the visual wake words dataset converted to ppm format.
The build command is:
make clean all
To run on GVSOC
make run platform=gvsoc
And on a connected GAPUINO or other GAP development board:
make run
This demo is also meant to be ran with camera collected frames in realtime. To run on Gapuino board with images grubbed from a real camera and show t, you need a special camera adaptor for the galaxycore color camera, and a modification on the board to correcty power supply the camera, the output is shown to a TFT adafruit LCD screen . Write an email to [email protected] to have more info on how to get the hardware.
Once you have the hardware you can uncomment the following flags in the makefile:
## Mute printf in source code
SILENT=1
## Enable image grub from camera and disaply output to lcd
FROM_CAMERA=1
and then type a:
make clean all run