tfjs-to-tf

tfjs-to-tf copied to clipboard

unable to convert blazepose 3d https://www.kaggle.com/models/mediapipe/blazepose-3d

getting below error tfjs_graph_converter ./model.json ./savedmodel --output_format tf_saved_model TensorFlow.js Graph Model Converter

Graph model: ./model.json Output: ./savedmodel Target format: tf_saved_model

Converting.... Error: 34 nodes in a cycle

I'm looking at the issue and let you know as soon as I find something.

@krishwin I could not reproduce the issue:

I was able to convert all 4 models successfully and without issue. What OS and which version did you use? Just for reference, here's the versions I tested with:

>tfjs_graph_converter --version

tfjs_graph_converter 1.6.2

Dependency versions:
    tensorflow 2.8.0
    tensorflowjs 3.18.0

@patlevin Thanks for looking into this. i am using version 1.6.1 tfjs_graph_converter 1.6.1

Dependency versions: tensorflow 2.6.2 tensorflowjs 3.18.0

i am using an intel mac and trying to upgrade to 1.6.2 or 1.6.3 , i have installed tensorflow-macos and tfjs_graph_converter 1.6.2. But getting below error . looks like tensorflow-macos is not compatible with intel Macs.

F tensorflow/c/c_api_experimental.cc:739] Non-OK-status: tensorflow::RegisterPluggableDevicePlugin(lib_handle->lib_handle) status: FAILED_PRECONDITION: 'host_callback' field in SP_StreamExecutor must be set. zsh: abort tfjs_graph_converter ./model.json ./savedmodel --output_format tf_saved_model

You could try using the tensorflow-cpu package instead - the problem seems to be due the TF version and since the converter doesn't use (or require) any GPU acceleration, that package should work.

Unfortunately I don't own any Apple hardware and the only Macs I could borrow are M1/M2 machines, so I can't really test this.