Keras input shape ValueError

Question

I have a problem and at the same time a question. I want to make an image classifier with Keras using Theano as Backend and a Sequential model.

>>> keras.__version__
'2.0.1'  
>>> theano.__version__
'0.9.0'

My input shape: INPUT_SHAPE = (3, 28, 28) #depth, size, size

Let's come to my problem. If I run my script at Windows 7 32 Bit, it gives me the error below out:

ValueError: ('The specified size contains a dimension with value <= 0', (-1024, 512))

If run it with the the input shape: INPUT_SHAPE = (28, 28, 3) #size, size, depth
It gives me out this error below:

ValueError: Error when checking model input: expected conv2d_1_input to have shape (None, 48, 48, 3) but got array with shape (1000, 3, 48, 48)

If I run the code on Elementary OS 64 Bit, it runs without any problems (INPUT_SHAPE = (3, 28, 28)).

My keras.json file for windows is:

{
  "backend": "theano",
  "epsilon": 1e-07,
  "floatx": "float32",
  "image_dim_ordering": "tf"
}

So, my question is: Is there such a big difference between different Operating Systems or where is my mistake? Just to remind, I used exactly the same code for both systems.

Aniket Bhatnagar · Accepted Answer · 2017-07-28 05:55:57Z

16

If your problem is not solved yet try using: from keras import backend as K K.set_image_dim_ordering('th') This would work good if you wish to use theano backend and have to use channels first configuration for image dimension ordering.

answered Jul 28, 2017 at 5:55

Aniket Bhatnagar

1761 silver badge3 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

maz · Accepted Answer · 2017-03-22 21:13:57Z

4

The problem you are having has to do with the expected dimension ordering.

Tensorflow ordering (tf): Shapes are expected to be (size_lines,size_columns,channel)
Theano ordering (th): Shapes are expected to be (channel,size_lines,size_columns)

If you change the ordering line in the keras.json file to "image_dim_ordering": "th" it should work. (i'd bet that's what's in your Elementary OS keras.json).

answered Mar 22, 2017 at 21:13

maz

1,98015 silver badges17 bronze badges

1 Comment

Dragonblf Over a year ago

Thanks for the answer and sorry for my late response. You're still right, but I recognized that my .keras file doesn't change anything. I installed keras without any envs and one time trought anaconda. After setting the image_data_format to 'channels_first' in my code it works without any issues. So it was totally my fault. :D But thanks for your time.

lmiguelmh · Accepted Answer · 2017-08-26 05:20:38Z

For switching to another backend you can change the configuration file located in:

Linux: $HOME/.keras/keras.json
Windows: %USER_PROFILE%/.keras/keras.json

This is the keras.json file for theano backend:

{
    "floatx": "float32",
    "epsilon": 1e-07,
    "backend": "theano",
    "image_data_format": "channels_first"
}

This is the keras.json file for tensorflow backend:

{
    "floatx": "float32",
    "epsilon": 1e-07,
    "backend": "tensorflow",
    "image_data_format": "channels_last"
}

This is what the documentation https://keras.io/backend/ says about the image_data_format property:

image_data_format: string, either "channels_last" or "channels_first". It specifies which data format convention Keras will follow. (keras.backend.image_data_format() returns it.)

For 2D data (e.g. image), "channels_last" assumes (rows, cols, channels) while "channels_first" assumes (channels, rows, cols).

For 3D data, "channels_last" assumes (conv_dim1, conv_dim2, conv_dim3, channels) while "channels_first" assumes (channels, conv_dim1, conv_dim2, conv_dim3).

Collectives™ on Stack Overflow

Keras input shape ValueError

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related