Implementation of Tensorflow.js using a simple static webpage and a custom ML model. An overview of the project can be found on YouTube.

Download the files, then from the command line, create a local server:

python -m http.server

Navigate to http://localhost:8000/index.html and http://localhost:8000/staticimage_index.html to see the resulting demos. For the webcam demo, you will be asked to share your webcam.

For the creation of the model, I used the Microsoft Kaggle Cats and Dogs Dataset. This dataset contains 12,500 cat images and 12,500 dog images.

Only a subset of the data was used to create the model (1,500 of each class). Google Cloud's AutoML service was used to train the model then it was exported to files to be used in Tensorflow.js.

The first demo in index.html uses the exported model to make predictions about five images. This is a pretty standard demo that shows the power of Tensorflow.js to make predictions at the time of image load.

In the second demo, located in webcam_index.js, uses the webcam to make predictions. This demo is only useful if you can point the camera to a nearby cat or dog. However, this shows how you could make continuous predictions using Tensorflow.js using the webcam as an edge device.