BatchNormalization Operation in TFLite

I converted a keras model to tflite which includes batch normalization layer.

Opening the tflite file in Netron, the batch normalization operation is separated into 2 operations of multiplication and addition.

When doing inference on a couple of test samples with tflite , the values are not just multiplied and added in batch normalization layer. There seems to be more operations than just a simple multiplication and addition.

Does anyone know if tensorflow lite is applying any operation besides that when doing batch normalization?

It’s usually removed and expanded. Here you can find a non-exhaustive list of all such operations for which this behavior is applied.