How to solve problem:Didn't find op for builtin opcode 'SHAPE' version '1'. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?

#1

Hi
When I tried to modify “tflite-trained.h” file with another model in C++ mode, there encountered a problem on my embedded system:

image
image1109×174 8.53 KB

Here are my settings:
OS SYSTEM: WIN10
Tensorflow version: 2.1.0
tf-nightly:2.1.0
Model architecture:
image
image1150×409 31.9 KB

How to solve it?

#2

Hi @sherry,

We do not recommend modifying the files in the downloaded C++ library, as the source code and model are tightly coupled. As you have found, slight variations in versions used to produce a TFLite file could break the build process.

If you wish to modify a model, we recommend doing so in your Studio project (you can use Expert Mode to see the Keras code). Once you have trained the new model, you can download the updated C++ library from the Deployment section. This will ensure that all functions/variables are working within your library.

#3

OK! Thank you so much

1 Like