I am trying to load a pretrained keras fully connected neural network model into openMV. I have converted the model to tflite but I get the following error:
OSError: Didn’t find op for builtin opcode ‘LOGISTIC’ version ‘1’. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?
Any idea how I can fix this?
Thanks for the response, it seems to me that the problem might be because I use a sigmoid activation function. Is there an alternative way to make it work despite using sigmoid?
You need to change the model to not use that.
I see, another question is that if I use standardscaler() to preprocess some features before training the model, and I save the standardscaler model also in order to apply on any input to the model before making predictions/classification. The standardscaler is a .bin file, is there a way to load it also in openmv?
I don’t know what that is exactly. We just really support EdgeImpulse for using models. It’s very hard to get models not trained by them working on the MCU. TensorFlow lite for microcontrollers has a lot of land-mines in it.