Trying to get an Edge Impulse Vision model onto the web.

First as a WASM

This is the normal way I would show an Edge Impulse model for tinyML but on the web using my version of an Edge Impulse web viewer. Note: With sounds. Kind of cool

index-wasm.html
The example kind of badly detects heart shaped rocks. How well or badly it works it not the point.

Then with TensorflowJS

Converting the saved Tensorflow model to a TensforflowJS model here
Main work in progress github here

Best so far

forweb/index.html Final should go here

forweb/index-cool.html Just working on this one

Best working backup

forweb/index-last-working.html

Best simplifications

forweb/index-edge-impulse-to-tfjs-basic-graph-demo.html

forweb/index-edge-impulse-to-tfjs-basic-layers-demo.html

Raw data test

forweb/index-raw-test.html The results should be unknown 0% heart Rock 100%

Crashing after about 20 seconds and ~10000 un-disposed tensors

This issue was fixed using:

  
   tf.engine().startScope()
  
       // Tensor stuff
  
   tf.engine().endScope()

forweb/backups/index-crashing.html

Research possible code to adapt

forweb/model.json The present model.json file converted from a saved tensorflow model exported from edge impulse

index-just-load.html Just load any Layers or Graph Vision saved TFJS model.json

forweb/index-squeezenet2.html Might be the best starting point, but need to make the webcam work on apple phones

index-squeezenet.html

index-mobilenet.html

index-mnist.html

index-from-image-mobilenet.html

Index-touch.html

index-load.html Try to just load the model and test it against an image

...