-
Open up the demo
-
Wait for the mobilenet model to load; if it doesn't within a minute or so, make sure you're using chrome and have a webcam.
-
Add examples of letters by clicking and holding on the buttons that say "Add Sample". An image of your most recent sample should show on the button.
-
Once you have enough examples (100 of each should do), click train model.
-
Wait for the model to finish training; then press start! The program should highlight the letter it thinks you're signing.
This example shows you how to predict letters from a webcam using transfer learning.
This example is deploying a pretrained MobileNet model and training another model using an internal mobilenet activation to predict different letters from the webcam defined by the user.
This is a Pure Javascript implementation of: https://github.com/tensorflow/tfjs-examples/tree/master/webcam-transfer-learning. Thank you TensorFlow.js for your flexible and intuitive APIs.