Registering two statistical functions with name 'FusedBatchNormV3,flops'! (Previous registration was in register /usr/local/lib/python3.10/dist-packages/tensorflow/python/framework/registry.py:65)"

[rsarka34]

KeyError Traceback (most recent call last)
in <cell line: 4>()
2 from tensorflow.keras.layers import Dense, Flatten, Conv2D, MaxPooling2D, Dropout
3
----> 4 from keras_flops import get_flops
5
6 # build model

4 frames
/usr/local/lib/python3.10/dist-packages/tensorflow/python/framework/registry.py in register(self, candidate, name)
55 if name in self._registry:
56 frame = self._registry[name][_LOCATION_TAG]
---> 57 raise KeyError(
58 "Registering two %s with name '%s'! "
59 "(Previous registration was in %s %s:%d)" %

KeyError: "Registering two statistical functions with name 'FusedBatchNormV3,flops'! (Previous registration was in register /usr/local/lib/python3.10/dist-packages/tensorflow/python/framework/registry.py:65)"

[m-shahpouri]

Hello, my friend. According to this answer #17 (comment) you have to go to

/home/name/miniconda3/envs/tf-2.13/lib/python3.9/site-packages/keras_flops/flops_registory.py

directory and comment or remove following lines:

@ops.RegisterStatistics("FusedBatchNormV3", "flops")
@ops.RegisterStatistics("Max", "flops")
@ops.RegisterStatistics("AddV2", "flops")

I hope it helps you.

Maybe better solution

However, I think this code is easier to use instead of editing a package file:

import tensorflow as tf
from tensorflow.python.profiler.model_analyzer import profile
from tensorflow.python.profiler.option_builder import ProfileOptionBuilder

def get_flops(model):
  forward_pass = tf.function(model.call, input_signature=[tf.TensorSpec(shape=(1,) + model.input_shape[1:])])
  graph_info = profile(forward_pass.get_concrete_function().graph, options=ProfileOptionBuilder.float_operation())
  flops = graph_info.total_float_ops
  return flops

HEIGHT = 224
WIDTH = 224
CHANNELS = 3

model = tf.keras.Sequential([
    tf.keras.Input(shape=(HEIGHT, WIDTH, CHANNELS), name='model_input'),
    tf.keras.layers.Conv2D(filters=3, kernel_size=(1, 1), padding='same', groups=3, activation='linear', name='DepthwiseConv2D_1'),
    tf.keras.layers.Dense(units=1, activation='relu', name='PointwiseConv2D_1')
    ])

# model.compile(optimizer='adam', loss='bce', metrics=['accuracy'])

flops = get_flops(model)
macs = flops / 2
print(f"MACs: {macs:,}")
print(f"FLOPs: {flops:,}")

MACs: 250,880
FLOPs: 501,760

Source: tensorflow/tensorflow#32809 (comment)

Just keep in mind that the source computes MACs tensorflow/tensorflow#32809 (comment)