tflite instance norm generation python

inst_norm.py

## tested on tf 2.0

from tensorflow import keras
import tensorflow as tf
from tensorflow.keras import layers
import tensorflow_addons as tfa

inputs = keras.Input(shape=(784,), name='digits')
x = tfa.layers.normalizations.InstanceNormalization()(inputs)
outputs = layers.Dense(10, activation='softmax', name='predictions')(x)

model = keras.Model(inputs=inputs, outputs=outputs, name='3_layer_mlp')
model.summary()
model.save('instance_norm.h5')

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
f = open('instance_norm.tflite', 'bw')
f.write(tflite_model)
f.close()

For broadcast binaryop :

from tensorflow import keras
import tensorflow as tf
from tensorflow.keras import layers
import tensorflow_addons as tfa

inputs = keras.Input(shape=(40,40,3), name='input')
x = tfa.layers.normalizations.InstanceNormalization()(inputs)
outputs = layers.Dense(10, activation='softmax', name='predictions')(x)

model = keras.Model(inputs=inputs, outputs=outputs, name='3_layer_mlp')
model.summary()
model.save('instance_norm.h5')

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
f = open('instance_norm.tflite', 'bw')
f.write(tflite_model)
f.close()

Thanks a lot!

Hi chunseoklee,

Did you test the converted model on device / using python test.
Even though the conversion works it breaks during inference due to instance normalization

Hi chunseoklee,

Did you test the converted model on device / using python test.
Even though the conversion works it breaks during inference due to instance normalization

I do not recall. If instance norm is split into base ops(like add, mul ...), i am sure that it will run on devices with the latest tflite interpreter.