Pi edition. Not production yet.
Unfortunately I misconfigured the train and am only just barely able to trust the masks and model as an instance segmentation model. There is no way I will be able to repair this in the time I have left so instead I have tried to work around it for the time being.
Make sure to export DISPLAY=:0 if there is no screen.
Bug: run without glsl shader and then with to get it to work ?!
Pi 8GB (Tested on Pi 4 Model B 8GB)
OpenNI2 capable device capable of both color and depth streams:
- Hacked Xbox Kinect
- Carmine 1.09 or 1.08 (Adjust image input size 320 x 240)
- Intel RealSense (Tested on) (Dependencies not yet included in dependencies.sh)
Edgetpu: Google Coral USB (Testing on)
Simply execute the dependencies.sh batch file
Make sure to set
framebuffer_width=1080
framebuffer_height=720
hdmi_force_hotplug=1
hdmi_group=1
hdmi_mode=16
hdmi_drive=2
in /boot/config.txt.
USB 3.0 Coral TPU
USB 3.0 Kinect OpenNI2 capable device
Pi ICE Tower CPU Cooling Fan
Pi 4 Model B Heat Sink Cooling Fan mounted on coral tpu.
Note: Use unthrottled edgetpu runtime at your own risk. Note: I am not liable for any of your damages. (Use at your own risk)