This demo shows how to perform transfer learning using the Layers API of TensorFlow.js.
It follows the procedure outlined in the Keras mnist_transfer_cnn example.
- A simple convnet was trained in Python Keras on only the first 5 digits [0..4] from the MNIST dataset. The resulting model is hosted at a URL and loaded into TensorFlow.js using
tf.loadLayersModel()
. - The convolutional layers are frozen, and the dense layers are fine-tuned in the browser to classify the digits [5..9].
To launch the demo, do
yarn
yarn watch