Project to deploy a Machine Learning Model on a Rasberry Pi
Machine learning based on tensorflow and a costum dataset & convert it to Lite model to run on embedded devices as raspberry pi or android
- clone this repository with
git clone --recurse-submodules -j8 https://github.com/Yeshey/SignLanguageDetection_TenserFlow_RaspberryPi.git - follow the
SignLanguageDetection.ipynbfile to setup, train and run/convert the model
- American sign language letters image
- See Nvidia available memory in linux with
watch -n 0.1 nvidia-smi - Activate environment `conda activate signLanguageDetector'
- RasberryPi Credentials:
- User: pi
- Passwd: raspberry
- Wifi Connection:
- Connect using the hostname:
ssh pi@raspberrypi - Connecting using ip (connected to yeshey-hotspot): 'ssh [email protected]'
- Commands:
sudo nano /etc/wpa_supplicant/wpa_supplicant.conf- file that says wich networks it will connect tosudo iwlist wlan0 scanning | grep ESSID- show available wlan connectionsiwconfig | grep wlan0- show currently active wlan connection
- Connect using the hostname:
- Check Teams
-
For Everyone doing an object detection: I recommend using this API for Training A very good tutorial from start to finish can be found here: https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html General This also works in the ICC
-
General for everyone wanting to annotate images, I can recommend https://supervise.ly/
-
General Tutorial for image classification: https://www.tensorflow.org/hub/tutorials/tf2_image_retraining
-
- Get DataSet of faces (like from https://thispersondoesnotexist.com)
- Or find datasets in this site for example: paperswithcode.com
- Label your dataset with something like https://github.com/tzutalin/labelImg
- Train Model in your home PC GPU
- check the links and info in MS Teams to do that
- Convert the model to tensorflow lite model, to run on your Coral TPU
- Or search for how to make it run on Coral TPU Stick
- Load it to rasberry Pi and make it work somehow.
- it has camera, Coral TPU, Keyboard.