Cannot import name relu6

[tensor-flower]

Using TensorFlow backend.
Traceback (most recent call last):
  File "train.py", line 9, in <module>
    from mobilenet_v2 import MobileNetv2
  File "/home/wq/MobileNetV2/mobilenet_v2.py", line 13, in <module>
    from keras.applications.mobilenet import relu6, DepthwiseConv2D
ImportError: cannot import name relu6

There seems to be some name changes since you write the script.

[gaoshine]

yes, if you use keras2.2 .
from keras.applications.mobilenet import relu6, DepthwiseConv2D .
i think in 2.2, it change something.

[EggyDream]

I think in keras mobilenet implement are moving place,
from keras.applications to keras_applications.

Personally I replace,
from keras.applications.mobilenet import relu6, DepthwiseConv2D
to,
from keras_applications.mobilenet import relu6
from keras.layers import DepthwiseConv2D

The reason I uese DepthswiseConv2D from layers
is I found that the mobilenet's DepthswiseConv2D is import from layers

Also It seems keras 2.2 has its own mobilenet-v2 implement.

[jmherrmann]

Had the same issue, this worked for me:

from keras import models
from keras_applications import mobilenet_v2

model = models.load_model(path_model_checkpoint, custom_objects={'relu6': mobilenet_v2.relu6})
model.summary()

Had the idea for the fix from here: keras-team/keras@468f080

[JustinAsdz]

@jmherrmann Still not work

AttributeError: module 'keras_applications.mobilenet_v2' has no attribute 'relu6'

can you help me with it?

[jmherrmann]

Which Keras version are you on @BokaiLIAN?

[JustinAsdz]

I have solved the problem by reinstall keras=2.1.4
Still thanks a lot @jmherrmann

[chxy95]

In keras 2.2, you can open the mobilenet_v2.py file in the keras_applications folder to find how to use relu6 rightly. In fact, you only need replace the Activation('relu6')(x) of mobilenet_v2.layers.ReLU(6.)(x).

[xincxiong]

As I reinstall keras=2.1.5, the backend is set to theano, but I cannot change the backend! why???
keras=2.1.5 still cannot work!!

Anyone solved this issue?? How ??

[SteveIb]

I have tried
with CustomObjectScope({'relu6': keras.layers.ReLU(6.),'DepthwiseConv2D': keras.layers.DepthwiseConv2D}):
model = load_model('****.hdf5')

but I got the following error:
ValueError: axes don't match array

my TF is 1.11 my keras is 2.2.4, python 2.7.
Im trying to convert the model on the same machine and environment i have trained on.
any suggestions?

[joseildofilho]

Why not use SeparableConv2D instead DepthConv2D and a relu with parameter max_value ?