Keras mobilenet can't be imported (missing Relu6 / DepthwiseConv2D) #38

luke14free · 2017-10-08T11:06:34Z

I have a custom trained MobileNet network from Keras and I bump into an issue about CoreML Tools not recognizing Relu6 as an activation function.

my keras model is something like:

import keras
from keras.layers.core import Flatten
initial_model = keras.applications.mobilenet.MobileNet(input_shape=(size, size, 3), include_top=False, weights='imagenet', classes=2)
last = initial_model.output
x = Flatten()(last)
preds = Dense(2, activation='softmax')(x)

model = Model(initial_model.input, preds)

And after training I try to convert it with:

from keras.utils.generic_utils import CustomObjectScope
with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}):
      convert = coremltools.converters.keras.convert("model.h5", input_names=['img'], image_input_names=['img'], class_labels=['class1', 'class2'])

But this raises:

RuntimeError: Unsupported option activation=relu6 in layer Dense(conv1_relu)

Because coreML Tools doesn't know what DepthwiseConv2D and Relu6 are.

The text was updated successfully, but these errors were encountered:

luke14free · 2017-10-10T14:10:38Z

Actually the only missing portion is the ReLU6, the DepthwiseConv2D is already there.

slin07 · 2017-10-10T19:32:37Z

PR#44 is up, should fix this issue.

luke14free · 2017-10-11T08:20:16Z

@slin07 yes! I tested with #44 and I am properly able to convert the model and it works quite well in my dummy frame categorizer app. 👍

slin07 · 2017-10-11T19:25:38Z

PR #44 has been merged. Closing this issue.

seantempesta · 2017-11-14T18:25:52Z

So I just ran into something similar:

ValueError: Unknown activation function:relu6

And I just built from source. @luke14free did you have to do anything else to convert your model?

slin07 · 2017-11-14T23:48:55Z

Can you provide (1) trace stack for this error and (2) the Keras version you're using?

luke14free · 2017-11-15T08:14:47Z

@seantempesta are you doing this?

from keras.utils.generic_utils import CustomObjectScope
with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}):
     ...

seantempesta · 2017-11-15T13:46:39Z

Full stack trace below. I was loading just the keras .h5 file.

(pythonenv) bash-3.2$ python coreml.py 
WARNING:root:Keras version 2.0.8 detected. Last version known to be fully compatible of Keras is 2.0.6 .
WARNING:root:TensorFlow version 1.4.0 detected. Last version known to be fully compatible is 1.2.1 .
Traceback (most recent call last):
  File "coreml", line 10, in <module>
    image_scale=1/255.)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/coremltools/converters/keras/_keras_converter.py", line 505, in convert
    predicted_probabilities_output = predicted_probabilities_output)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/coremltools/converters/keras/_keras2_converter.py", line 161, in _convert
    model = _keras.models.load_model(model)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/models.py", line 239, in load_model
    model = model_from_config(model_config, custom_objects=custom_objects)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/models.py", line 313, in model_from_config
    return layer_module.deserialize(config, custom_objects=custom_objects)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/layers/__init__.py", line 54, in deserialize
    printable_module_name='layer')
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/utils/generic_utils.py", line 139, in deserialize_keras_object
    list(custom_objects.items())))
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/engine/topology.py", line 2487, in from_config
    process_layer(layer_data)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/engine/topology.py", line 2473, in process_layer
    custom_objects=custom_objects)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/layers/__init__.py", line 54, in deserialize
    printable_module_name='layer')
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/utils/generic_utils.py", line 141, in deserialize_keras_object
    return cls.from_config(config['config'])
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/engine/topology.py", line 1252, in from_config
    return cls(**config)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/layers/core.py", line 283, in __init__
    self.activation = activations.get(activation)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/activations.py", line 95, in get
    return deserialize(identifier)
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/activations.py", line 87, in deserialize
    printable_module_name='activation function')
  File "/Users/sean/src/tmp/Corellian/CoreML/mlvirtualenv/pythonenv/lib/python2.7/site-packages/keras/utils/generic_utils.py", line 159, in deserialize_keras_object
    ':' + function_name)
ValueError: Unknown activation function:relu6

seantempesta · 2017-11-15T13:50:25Z

I ended up completing the conversion by just converting the model after building it in Keras and loading the weights. I can't post the entire file, but here's the gist:

##############################################################
############      LET'S DEFINE OUR MODEL      #################
###############################################################
modelBase = applications.MobileNet(weights='imagenet',include_top=False, input_shape = (img_width, img_height, 3))
...
model.load_weights('mobilenet-weights.h5')
import coremltools
coreml_model = coremltools.converters.keras.convert(model)

lahayeg · 2017-11-27T09:55:03Z

Hi @seantempesta does not seems to work for me. Please provide more sample if possible

greis · 2017-11-29T05:35:32Z

This seems to fix the issue for me keras-team/keras#7431 (comment)

tatsuya-ogawa · 2017-12-04T13:24:56Z

@seantempesta You are probably using coremltools from pip.
It seems that it has not been released yet so you need to build it yourself.

seantempesta · 2017-12-04T19:01:18Z

@tatsuya-ogawa: No, I built it from source. Granted that was 19 days ago, so I don't know if you've merged a fix recently?

bkou · 2018-03-27T22:14:45Z

If your keras is at "from tensorflow.python import keras" instead of at "import keras" then here is what worked for me:

from tensorflow.python.keras._impl.keras.utils.generic_utils import CustomObjectScope
from tensorflow.python.keras._impl.keras.applications import mobilenet
from tensorflow.python.keras._impl.keras.models import load_model
with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):
    model = load_model('weights.hdf5')

schliffen · 2018-09-24T16:27:17Z

with tensorflow 1.9 and keras 2.2.0 this is what I did and it worked completely:
from tensorflow.python.keras.models import load_model
from keras.applications.mobilenet import MobileNet
from tensorflow.python.keras.applications import mobilenet
from tensorflow.python.keras.utils.generic_utils import CustomObjectScope
with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):
loaded_model.load_weights('model_256.h5')

for complete model you should import your model separately, e.g. using model_from_json

SteveIb · 2018-10-25T12:31:16Z

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?

schliffen · 2018-10-26T08:17:25Z

I suggest for conversion use Keras 2.1.6 to import relu6- import relu6 from "keras.applications.mobilenet". the rest is just as I explained above. (tensorflow version does not matter)

SteveIb · 2018-10-28T08:02:43Z

I suggest for conversion use Keras 2.1.6 to import relu6- import relu6 from "keras.applications.mobilenet". the rest is just as I explained above. (tensorflow version does not matter)

Hi, i downgraded keras to 2.1.6

My code:

import keras

from keras.models import model_from_json

with open('model.json','r') as f:
json = f.read()
loaded_model = model_from_json(json)

from tensorflow.python.keras.models import load_model
from keras.applications.mobilenet import MobileNet
from tensorflow.python.keras.applications import mobilenet
from tensorflow.python.keras.utils.generic_utils import CustomObjectScope
with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}):
loaded_model.load_weights('**********.hdf5')

Im getting the following error:

ValueError: Unknown layer: ReLU

schliffen · 2018-10-28T08:42:30Z

You should train it again, with this version

…

On Oct 28, 2018 10:02 AM, "SteveIb" ***@***.***> wrote: I suggest for conversion use Keras 2.1.6 to import relu6- import relu6 from "keras.applications.mobilenet". the rest is just as I explained above. (tensorflow version does not matter) Hi, i downgraded keras to 2.1.6 My code: import keras from keras.models import model_from_json with open('model.json','r') as f: json = f.read() loaded_model = model_from_json(json) from tensorflow.python.keras.models import load_model from keras.applications.mobilenet import MobileNet from tensorflow.python.keras.applications import mobilenet from tensorflow.python.keras.utils.generic_utils import CustomObjectScope with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}): loaded_model.load_weights('**********.hdf5') Im getting the following error: ValueError: Unknown layer: ReLU — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#38 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADmxxrE7eeFO5AtZZZeCvZPnRmx3HzC1ks5upWSlgaJpZM4Pxqxu> .

SteveIb · 2018-10-28T10:19:52Z

You should train it again, with this version
…
On Oct 28, 2018 10:02 AM, "SteveIb" @.> wrote: I suggest for conversion use Keras 2.1.6 to import relu6- import relu6 from "keras.applications.mobilenet". the rest is just as I explained above. (tensorflow version does not matter) Hi, i downgraded keras to 2.1.6 My code: import keras from keras.models import model_from_json with open('model.json','r') as f: json = f.read() loaded_model = model_from_json(json) from tensorflow.python.keras.models import load_model from keras.applications.mobilenet import MobileNet from tensorflow.python.keras.applications import mobilenet from tensorflow.python.keras.utils.generic_utils import CustomObjectScope with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}): loaded_model.load_weights('*******.hdf5') Im getting the following error: ValueError: Unknown layer: ReLU — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#38 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/ADmxxrE7eeFO5AtZZZeCvZPnRmx3HzC1ks5upWSlgaJpZM4Pxqxu .

I trained again.
File "toTfLite.py", line 26, in
loaded_model = model_from_json(json)

ValueError: Unknown activation function:relu6

SteveIb · 2018-10-28T10:26:22Z

I'm using this library to train on MobileNet
https://github.com/experiencor/keras-yolo2

fasecity · 2018-12-29T21:03:49Z

Had this same issue: my fix:

from pathlib import Path #from tf.keras.models.model_from_json import model_from_json

# Load the json file that contains the model's structure
f = Path('D:\\Class\\mytflite\\Model_forker\\my_model.json')
model_structure = f.read_text()


from tensorflow.python.keras.utils.generic_utils import CustomObjectScope
from tensorflow.python.keras.applications import mobilenet
from tensorflow.python.keras.models import load_model
with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):
    # Recreate the Keras model object from the json data
    model = tf.keras.models.model_from_json(model_structure)
    # Re-load the model's trained weights
    model.load_weights('D:\Class\mytflite\Model_forker\my_model.h5')

prabhanjan215 · 2019-01-05T06:27:05Z

Am facing the issue still

from tensorflow.python.keras.models import load_model
from keras.applications.mobilenet import MobileNet
from tensorflow.python.keras.applications import mobilenet
from tensorflow.python.keras.utils.generic_utils import CustomObjectScope

with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):transfer_model = load_model('modeltransfer1.h5')

AttributeError: module 'tensorflow.python.keras.applications.mobilenet' has no attribute 'relu6'

Please help on this

fasecity · 2019-01-09T18:38:46Z

Am facing the issue still

from tensorflow.python.keras.models import load_model
from keras.applications.mobilenet import MobileNet
from tensorflow.python.keras.applications import mobilenet
from tensorflow.python.keras.utils.generic_utils import CustomObjectScope

with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):transfer_model = load_model('modeltransfer1.h5')

AttributeError: module 'tensorflow.python.keras.applications.mobilenet' has no attribute 'relu6'

Please help on this

what worked for me is loading the weights and json individually.
` ...
from tensorflow.python.keras.models import load_model
with CustomObjectScope({'relu6': mobilenet.relu6,'DepthwiseConv2D': mobilenet.DepthwiseConv2D}):
#------------------------here----------------------------------------------------

Recreate the Keras model object from the json data

model = tf.keras.models.model_from_json(model_structure) #<---- comes from the model.json
# Re-load the model's trained weights
model.load_weights('modeltransfer1_weihts.h5'')`

Unify attention modules

srikris assigned slin07 Oct 8, 2017

srikris added the bug Unexpected behaviour that should be corrected (type) label Oct 8, 2017

slin07 closed this as completed Oct 11, 2017

mrgloom mentioned this issue Oct 18, 2018

Retrained MobileNet and failed to load it for prediction keras-team/keras#7431

Closed

Birch-san pushed a commit to Birch-san/coremltools that referenced this issue Nov 27, 2022

Merge pull request apple#38 from huggingface/one_attentino_module

e5d9baf

Unify attention modules

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Keras mobilenet can't be imported (missing Relu6 / DepthwiseConv2D) #38

Keras mobilenet can't be imported (missing Relu6 / DepthwiseConv2D) #38

luke14free commented Oct 8, 2017 •

edited

Loading

luke14free commented Oct 10, 2017

slin07 commented Oct 10, 2017

luke14free commented Oct 11, 2017

slin07 commented Oct 11, 2017

seantempesta commented Nov 14, 2017

slin07 commented Nov 14, 2017

luke14free commented Nov 15, 2017

seantempesta commented Nov 15, 2017

seantempesta commented Nov 15, 2017

lahayeg commented Nov 27, 2017

greis commented Nov 29, 2017

tatsuya-ogawa commented Dec 4, 2017

seantempesta commented Dec 4, 2017

bkou commented Mar 27, 2018 •

edited

Loading

schliffen commented Sep 24, 2018

SteveIb commented Oct 25, 2018

schliffen commented Oct 26, 2018

SteveIb commented Oct 28, 2018

schliffen commented Oct 28, 2018 via email

SteveIb commented Oct 28, 2018

SteveIb commented Oct 28, 2018

fasecity commented Dec 29, 2018

prabhanjan215 commented Jan 5, 2019

fasecity commented Jan 9, 2019

Keras mobilenet can't be imported (missing Relu6 / DepthwiseConv2D) #38

Keras mobilenet can't be imported (missing Relu6 / DepthwiseConv2D) #38

Comments

luke14free commented Oct 8, 2017 • edited Loading

luke14free commented Oct 10, 2017

slin07 commented Oct 10, 2017

luke14free commented Oct 11, 2017

slin07 commented Oct 11, 2017

seantempesta commented Nov 14, 2017

slin07 commented Nov 14, 2017

luke14free commented Nov 15, 2017

seantempesta commented Nov 15, 2017

seantempesta commented Nov 15, 2017

lahayeg commented Nov 27, 2017

greis commented Nov 29, 2017

tatsuya-ogawa commented Dec 4, 2017

seantempesta commented Dec 4, 2017

bkou commented Mar 27, 2018 • edited Loading

schliffen commented Sep 24, 2018

SteveIb commented Oct 25, 2018

schliffen commented Oct 26, 2018

SteveIb commented Oct 28, 2018

schliffen commented Oct 28, 2018 via email

SteveIb commented Oct 28, 2018

SteveIb commented Oct 28, 2018

fasecity commented Dec 29, 2018

prabhanjan215 commented Jan 5, 2019

fasecity commented Jan 9, 2019

Recreate the Keras model object from the json data

luke14free commented Oct 8, 2017 •

edited

Loading

bkou commented Mar 27, 2018 •

edited

Loading