-
Notifications
You must be signed in to change notification settings - Fork 0
Model Training Guidelines with Teachable Machine
Yehoon (Luke) Joo edited this page Feb 11, 2022
·
8 revisions
For consistency across our trained models, please follow the guidelines below!
Teachable Machine - Pose Project
- Go to the Pose Project page.
- Edit Class 1 to target
- Add more classes to capture possible user poses which are NOT the target pose. (e.g. idle, armsUp, armsDown, bendOver, ...)
- For data collection, click on Webcam.
- Click on the settings icon and set it to FPS: 24, Hold to Record: off, Delay: 5 seconds, Duration: 10 seconds, and click Save Settings.
- Click on Record 10 Seconds, and set yourself within 5 seconds (the delay time) to record your pose data for 10 seconds (the duration time). Ideally fit your entire body (head to toe) into the center of the camera. Moving around a little can help to add some variance to the model, making the model perform better for different user input.
- Once data collection is complete, click on Train Model with its default settings. Make sure to not change browser tabs in order to complete the training process.
- Once training is complete, you can test out your model realtime and see model predictions on the right.
- Click on Export Model, then Upload my model to generate a URL link for your custom trained model.
- Copy the generated URL link to the Firebase database along with your reference pose image.
- Click on the menu tab to the top left, and click on Download project as a file.
-
Rename the downloaded file to your model's name (e.g. letter_p.tm).
-
You can load this model by clicking on Open project from file and selecting the downloaded .tm file.