Inception3 not supported ?

[MichalMisiaszek]

I uploaded custom tensorflowlite model which is based on InceptionV3,
plugin shows error:
The model is INCOMPATIBLE. It may contain unrecognized custom ops, or not FlatBuffer format: java.lang.IllegalArgumentException: Internal error: Cannot create interpreter: Didn't find op for builtin opcode 'CONV_2D' version '2'

[MichalMisiaszek]

It was my mistake, the options were invalid. But I fixed options and now I get error:
E/ModelDownloadManager( 4245): The model is incompatible with TFLite and the app is not upgraded, do not download
The model is 100% TFLLite I just uploaded it and used successfully on my computer.
My options are:
FirebaseModelInputOutputOptions options = FirebaseModelInputOutputOptions([ FirebaseModelIOOption(FirebaseModelDataType.BYTE, [1, 800, 800, 3]) ], [ FirebaseModelIOOption(FirebaseModelDataType.BYTE, [1, 8142]) ]);
Maybe I am missing something ?

[MichalMisiaszek]

I dowloaded model from Firebase to be sure it was not broken on the way and it is working in Python script on my Mac but not in plugin.

[azihsoyn]

Thank you for reporting the detailed issue!

[MichalMisiaszek]

I am trying to dig into that but I don't see where this exception is coming from.

[azihsoyn]

I'll check this issue within a week. But it may take time.

[MichalMisiaszek]

Sad to hear it. Your plugin is the only I found for custom models, I will try to find the issue and fork in in worse case.

[MichalMisiaszek]

Did you run your example recently ? Gradle fails for me without explanation.

[MichalMisiaszek]

I corrected another input/output format config and I am getting.

E/CustomCompatChecker(20300): The model is INCOMPATIBLE. It may contain unrecognized custom ops, or not FlatBuffer format: java.lang.IllegalArgumentException: Internal error: Cannot create interpreter: Didn't find op for builtin opcode 'CONV_2D' version '2'
E/CustomCompatChecker(20300): Registration failed.
E/ModelResourceManager(20300): Error preloading model resource
E/ModelResourceManager(20300): com.google.firebase.ml.common.FirebaseMLException: Remote model load failed with the model options: Local model name: unspecified. Remote model name: detector.
E/ModelResourceManager(20300): 	at com.google.android.gms.internal.firebase_ml.zzon.zza(Unknown Source:33)
E/ModelResourceManager(20300): 	at com.google.android.gms.internal.firebase_ml.zzpr.zza(Unknown Source:109)```

[MichalMisiaszek]

Hi,
I dig deeper and it seems a lot of people have problems with TF converted to TFLite and then used on Android. I have not seen a solution to the problem yet :(.

[MichalMisiaszek]

So it seems your example works fine, the issue is with more complex optimized models.
I will try complex but not optimized and see what will happen. I wish the exceptions would be more informative.

[azihsoyn]

I'll resolve #54 in next release.

Thanks.

[MichalMisiaszek]

Do you know what is the problem ?

[MichalMisiaszek]

Hi did you managed to solve it ? The same problem directly on Android it suppose to be solved by upgrading to newer Tensorflow Lite recently released.
tensorflow/tensorflow#28163