TypeError: unsupported operand type(s) for /: 'Dimension' and 'int'

[Yaredoh]

Am using tensorflow 1.0 and python 3.6. can you check it ?

Traceback (most recent call last):
File "train.py", line 253, in
main()
File "train.py", line 144, in main
net = DeepLabResNetModel({'data': image_batch}, is_training=args.is_training, num_classes=args.num_classes)
File "/home/batuser/Documents/color/kaffe/tensorflow/network.py", line 48, in init
self.setup(is_training, num_classes)
File "/home/batuser/Documents/color/deeplab_resnet/model.py", line 114, in setup
.atrous_conv(3, 3, 256, 2, padding='SAME', biased=False, relu=False, name='res4a_branch2b')
File "/home/batuser/Documents/color/kaffe/tensorflow/network.py", line 22, in layer_decorated
layer_output = op(self, layer_input, *args, **kwargs)
File "/home/batuser/Documents/color/kaffe/tensorflow/network.py", line 170, in atrous_conv
kernel = self.make_var('weights', shape=[k_h, k_w, c_i / group, c_o])
TypeError: unsupported operand type(s) for /: 'Dimension' and 'int'

[cbessega]

I had the same problem. You will need to do a bunch of small tweaks to make it work in Python 3.6.
In case you are not especially interested in Python 3.6, the repo runs perfectly fine in 2.7

In that particular case what happens is that get_shape() in a couple of places of kaffe/tensorflow/network.py returns a dimension.

To get the shape as a list of ints, do tensor.get_shape().as_list()

[Yaredoh]

Thanks, @cbessega I will check it on 2.7.

[NitishMutha]

Hi, @cbessega, @DrSleep can you let me know what all fixes would be required to make it compatible with 3.6?
Since i am on windows machine, tensorflow on windows just supports python >= 3.5, so i would have to make those fixes to make it work on my machine.

[yemanbh]

type casting the Dimension to int will solve the problem: if you have a=10, dim = X.shape[10], then instead of adim use aint(dim)

[fastlater]

For newbies (like me) and as mentioned in #80
from line:
kernel = self.make_var('weights', shape=[k_h, k_w, c_i / group, c_o])
change it to:
kernel = self.make_var('weights', shape=[k_h, k_w, int(c_i) / group, c_o])

[dwleeKAIST]

sz = input.shape
sz[0] : Dimension type.
You need to use typecasting by ' int(sz[0]) '.

[kourouklides]

FYI I had exactly the same error when running the Transformer tutorial and I managed to resolve it by adding .numpy() at the end, i.e. temp_learning_rate_schedule(tf.range(40000, dtype=np.float32)).numpy()