microsoft / onnxjs

ONNX.js: run ONNX models using JavaScript

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

TypeError: cannot resolve operator 'Min' with opsets: ai.onnx v12

DrhF opened this issue · comments

commented

Hello!
I'm running an app on React, with onnxjs v0.1.8. I have a tf.keras model converted to onnx v12.

When I'm running predictions, I have an error in log:

index.js?44d8:109 TypeError: cannot resolve operator 'Min' with opsets: ai.onnx v12
    at Object.e.resolveOperator (onnx.min.js?9bb0:1)
    at t.resolve (onnx.min.js?9bb0:14)
    at e.initializeOps (onnx.min.js?9bb0:14)
    at eval (onnx.min.js?9bb0:14)
    at t.event (onnx.min.js?9bb0:1)
    at e.initialize (onnx.min.js?9bb0:14)
    at eval (onnx.min.js?9bb0:14)
    at eval (onnx.min.js?9bb0:14)
    at Object.eval [as next] (onnx.min.js?9bb0:14)
    at a (onnx.min.js?9bb0:14)

Model architecture:

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
mobilenetv2_1.00_224 (Model) (None, 1280)              2257984   
_________________________________________________________________
flatten (Flatten)            (None, 1280)              0         
_________________________________________________________________
dense (Dense)                (None, 512)               655872    
_________________________________________________________________
dropout (Dropout)            (None, 512)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 9)                 4617      
=================================================================

While converting a model using onnxmltools, I got a warning:

The ONNX operator number change on the optimization: 477 -> 173

Got no errors while running pure mobilenetv2_1.00_224 onnx model in js app.

Could you help me, please, with this error, at least the origin of error: onnx.js issue, onnx converter issue or model issue?

The Min operator is not currently supported by onnx.js unfortunately 😿 https://github.com/microsoft/onnxjs/blob/master/docs/operators.md