Don't work with a project from lobe, but work from Google Teachable Machine only

[marcbuils]

Hi, I tested this project, but I can't genarate compatible files from MS Lobe. But I tested with an export from Google Teachable Machine and this works very well.

Steps :

• Export a project from Lobe
• Select the format "Tenserflow Saved Model"
• execute the following command to convert the export for TenserFlow js:

  tensorflowjs_converter --input_format=tf_saved_model --output_format=tfjs_graph_model . ./web-model

• copy / paste the exported files in le directory public/model

Actual result :

• an error message is display on the web page
• the file metadata.json doesn't exist
• the library "@teachablemachine/image" is used in the file Prediction.js

Expected result :

• The labels who correspond to my webcam are displayed

My usage of tensorflowjs_converter is may be bad ?

[KartoffelToby]

Same here, im confused.

[slava554]

it looks like teachablemachine/image requires model in tfjs_layers_model format, but I have not find any way to convert exported model in tf_saved_model to it. I have tried to use tfjs_graph_model with tensorflow.js directly, but have failed (Argument 'x' passed to 'gather' must be numeric tensor, but got string tensor).

[KartoffelToby]

i can reproduce this error, with tfjs-node and the direct saved model its kind of work but i got wrong results.

i requested allready a tensorflow js Export direct from the lobe App #14

[adammenges]

Thanks for this! We're looking into what's going on now. 😄

[ohjs88]

Is there an update on this?

[mbeissinger]

Fixed with #17