PRELU operation error

dbinaghi — Mon, 10 Jan 2022 18:12:48 GMT

Hi,

I'm using a iMX8M Plus with a quantised tensorflow lite model (uint8).
The tflite runtime version is the 2.3.1.

My model contains some PRELU operations that according to the guide is a supported operation.

A warning is reported:

WARNING: Operator PRELU (v1) refused by NNAPI delegate: Input should be Float or UINT8

I think this is due to the fact that the tensor during processing becomes an int8 but it should be supported. This warning does not happen if I use a RELU activation function instead of PRELU.

The quantised model is attached.

Could you help me?
Thanks!

eIQ Machine Learning SoftwareのトピックPRELU operation error

PRELU operation error