TFJS DeepLab Porting Error

To get help from the community, we encourage using Stack Overflow and the tensorflow.js tag.

TensorFlow.js version

Python TFJS: 0.8.5 NPM TFJS: 1.0.0

Browser version

Chrome latest

Describe the problem or feature request

Hi, I am trying to port a pretrained DeepLab model to TFJS.

The model is trained in Python and saved as a tf_frozen_model. I convert it to a TFJS model using an tfjs-converter 0.8.5.

I load this converted model to a React app with TFJS version 1.0.0 with loadGraphModel.

Running model.predict(<tf.Tensor4D>) with this model throws the following error:

Uncaught (in promise) TypeError: Cannot read property 'backend' of undefined
at t.readSync
at t.moveData
at t.get
at t.reshape
at Ot.runKernel.$x
at t.scopedRun
at t.runKernel
at reshape_
at reshape

Any suggestions on how to fix this would be greatly appreciated. Thanks!

Code to reproduce the bug / link to feature request

[aptlin]

@kavikode, you might take a look at https://github.com/tensorflow/tfjs-models/pull/229, which has a lot of issues at the moment and will change a lot in the next few days. If you want to dive into the code, you might follow the instructions in the demo README.

[dhrumil83]

Does your model have any .pb files? In that case you need to convert the model once more since its the old format and not valid for tfjs 1.0.

https://github.com/tensorflow/tfjs-converter

I have a model converted with a previous version of TensorFlow.js converter (0.15.x), that is in .pb format. How do I convert it to the new JSON format?

I am looking forward to the next piece of computer vision(segmentation) optimized to work in the browser. Body pix example is great, but need ability to do that on generic objects as well. Hopefully some more examples/models related to this coming soon.