Closed Rahn80643 closed 2 years ago
Hi @Rahn80643
When a dimension is unspecified or unknown (usually it is the batch size), we need to infer the size across the whole model. We have two "modes": ValidateOnly(the one by default) and InferOutputShape.
could you try adding to your command when calling Arm NN --infer-output-shape
?
could you have a look to issue #603 and see if it helps you
Thank you Kind regards
Hi @TeresaARM
Thank you for providing the issue thread, I can execute the command successfully with --infer-output-shape
and the inference code could be executed successfully when setting m_InferAndValidate = true;
as well.
Thank you!
Best Regards, Rahn
Hi @Rahn80643
glad to hear your model is working now!
Hi,
I'm trying to apply a BiSeNetV2 model implemented by TensorFlow2 API to Arm NN, but I met the following error message:
Fatal: Armnn Error: Unspecified dimension while using ShapeInferenceMethod::ValidateOnly
The tflite model is converted based on the tutorial. The followings are the settings of the environment:
How to know which layer causes the error above, and is the error above related to dimension of layers?
The attachment contains the tflite model I tested on Arm NN. TF2_tflite_test.tar.gz
Best Regards, Rahn