tensorflow: Convolutional layers cannot be used multiple times

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): no
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
• TensorFlow installed from (source or binary): source (branch 1.4)
• TensorFlow version (use command below): 1.4.0-dev
• Python version: 3.5
• Bazel version (if compiling from source): 5.4.0
• CUDA/cuDNN version: 8.0/6.0
• GPU model and memory: nVidia 1080Ti 11G
• Exact command to reproduce: run the script below

Describe the problem

Keras convolutional layers cannot be used multiple times without creating a name conflict. This is especially bad when trying to copy layers from one model to another (See the second example below). This was working a few weeks ago. Here is a simple test case that used to work:

#!/usr/bin/env python

import tensorflow as tf

from tensorflow.contrib.keras.api.keras.layers import Conv2D
from tensorflow.contrib.keras.api.keras.layers import Input

a = Input(shape=(None, 32, 3))
c = Conv2D(32, (3, 3))
c(a)
c(a)

I added some print statement to different versions of the code. It looks like the convolutional ops are created with the following names in the old code:

1. conv2d/convolution/
2. conv2d/convolution_1/

In the new code they are:

1. conv2d/convolution/
2. conv2d/convolution/

It looks like the name scope code was changed recently. I specifically notice that some of the scope handling was moved to init where it used to happen when the function was called.

This is a slightly more complex version of the code that shows that copying convolutional layers doesn’t work:

#!/usr/bin/env python

import tensorflow as tf

from tensorflow.contrib.keras.api.keras.layers import Conv2D
from tensorflow.contrib.keras.api.keras.layers import Input
from tensorflow.contrib.keras.api.keras.models import Model

a = Input(shape=(None, 32, 3))
b = a
b = Conv2D(32, (3, 3))(b)
mod1 = Model(inputs=a, outputs=b)
a = Input(shape=(None, 32, 3))
b = a
for layer in mod1.layers[1:]:
    b = layer(b)
mod2 = Model(inputs=a, outputs=b)

This is an example of the traceback:

Traceback (most recent call last):
  File "/root/vish/test.py", line 16, in <module>
    b = layer(b)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/keras/_impl/keras/engine/topology.py", line 252, in __call__
    output = super(Layer, self).__call__(inputs, **kwargs)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/layers/base.py", line 577, in __call__
    outputs = self.call(inputs, *args, **kwargs)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/layers/convolutional.py", line 172, in call
    outputs = self._convolution_op(inputs, self.kernel)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/ops/nn_ops.py", line 841, in __call__
    return self.conv_op(inp, filter)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/ops/nn_ops.py", line 503, in __call__
    return self.call(inp, filter)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/ops/nn_ops.py", line 191, in __call__
    name=self.name)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 631, in conv2d
    data_format=data_format, name=name)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2959, in create_op
    self._add_op(ret)
  File "/opt/rh/rh-python35/root/usr/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2599, in _add_op
    "is already used" % op.name)
ValueError: cannot add op with name conv2d/convolution as that name is already used