Custom model failed to run

[ShigemoriHakura]

When init the model using BaseImagePredictor It returns:

Version: 2.11.0 Input [0]: name: serving_default_input:0, type: Float32, dimensions: [1,256,256,3], quantizationParams: {scale: 0 zeroPoint: 0} Output [0]: name: StatefulPartitionedCall:0, type: Float32, dimensions: [1,42], quantizationParams: {scale: 0 zeroPoint: 0} Output [1]: name: StatefulPartitionedCall:2, type: Float32, dimensions: [1,20], quantizationParams: {scale: 0 zeroPoint: 0} Output [2]: name: StatefulPartitionedCall:1, type: Float32, dimensions: [1,1], quantizationParams: {scale: 0 zeroPoint: 0}

TFLite Warning: Encountered unresolved custom op: %s.

TFLite Warning: Node number %d (%s) %s.

Exception: Ops not found.

I'm not sure which ops is needed as the debug info don't give me much information.

model_ten.zip

[asus4]

@ShigemoriHakura Your model contains custom ops not supported in the default TFLite binary. Or using extended libraries such as select-ops.

(We should parse %s in the log message, sorry for the inconvenience)

[ShigemoriHakura]

hmmm ok, so do you know which custom ops is needed? (or how can I know the %s is what)

[asus4]

The code is here. The parsing va_list is disabled as it causes a crash randomly.

https://github.com/asus4/tf-lite-unity-sample/blob/66985ac9836565b23e61c104f2fd589b748fab6a/Packages/com.github.asus4.tflite/Runtime/ErrorReporter.cs#L35-L50

Here are the list built-in ops in TensorFlow Lite ↓ https://github.com/asus4/tf-lite-unity-sample/blob/master/Packages/com.github.asus4.tflite/Runtime/BuiltinOps.cs

[ShigemoriHakura]

The code is here. The parsing va_list is disabled as it causes a crash randomly.

https://github.com/asus4/tf-lite-unity-sample/blob/66985ac9836565b23e61c104f2fd589b748fab6a/Packages/com.github.asus4.tflite/Runtime/ErrorReporter.cs#L35-L50

Here are the list built-in ops in TensorFlow Lite ↓ https://github.com/asus4/tf-lite-unity-sample/blob/master/Packages/com.github.asus4.tflite/Runtime/BuiltinOps.cs

Thx, I founded the missing op is Landmarks2TransformMatrix, which is included in MediaPipe https://github.com/google/mediapipe/blob/master/mediapipe/util/tflite/operations/landmarks_to_transform_matrix.cc So I have to compile the tensorflow binary myself?

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.