The 1st entry of padding argument must be a tuple of 2 integers. Received: 3

[gohar94]

ubuntu@ip-10-0-0-5:~/keras/keras/caffe$ python caffe2keras.py -load_path 'models/' -prototxt 'deploy_new.prototxt' -caffemodel 'go.caffemodel'
Using Theano backend.
Converting model...
CREATING MODEL
Traceback (most recent call last):
  File "caffe2keras.py", line 46, in <module>
    main(args)
  File "caffe2keras.py", line 34, in main
    model = convert.caffe_to_keras(args.load_path+'/'+args.prototxt, args.load_path+'/'+args.caffemodel, debug=args.debug)
  File "/home/ubuntu/keras/keras/caffe/convert.py", line 53, in caffe_to_keras
    debug)
  File "/home/ubuntu/keras/keras/caffe/convert.py", line 197, in create_model
    input_layers = ZeroPadding2D(padding=(pad_h, pad_w), name=name + '_zeropadding')(input_layers)
  File "/home/ubuntu/keras/keras/legacy/interfaces.py", line 87, in wrapper
    return func(*args, **kwargs)
  File "/home/ubuntu/keras/keras/layers/convolutional.py", line 1641, in __init__
    '1st entry of padding')
  File "/home/ubuntu/keras/keras/utils/conv_utils.py", line 30, in normalize_tuple
    str(n) + ' integers. Received: ' + str(value))
ValueError: The `1st entry of padding` argument must be a tuple of 2 integers. Received: 3

I keep getting the error above - anyone know why this might be happening?

[qxcv]

That's really odd. What happens if you edit /home/ubuntu/keras/keras/utils/conv_utils.py so that normalize_tuple prints out its argument type? Something like this:

def normalize_tuple(value, n, name):
    """
    … docstring snipped…
    """
    # v should print "int" and an int-ish value, unless there's something weird going on.
    print(type(value))
    print(value)
    # ^
    if isinstance(value, int):
        return (value,) * n
    else:
        try:
            value_tuple = … # rest of function unchanged

My guess is that there's a numpy integer floating around somewhere in there, and the check in at the beginning of normalize_tuple is failing to spot it.

[gohar94]

<type 'long'>
3

This is what it prints!

[gohar94]

The number of values in the tuple seems to be an issue, the type is long which seems to be acceptable. Any ideas why this might be happening? Thanks a bunch for the help.

[qxcv]

isinstance(long(3), int) should be False, which would cause the observed problem. Running it with Python 3 (which has no long) or replacing isinstance(n, int) with isinstance(n, (int, long)) (on the Keras side) will hopefully make it work.

[gohar94]

Thanks, that resolved the error. Getting this now, though:

Traceback (most recent call last):
  File "caffe2keras.py", line 46, in <module>
    main(args)
  File "caffe2keras.py", line 34, in main
    model = convert.caffe_to_keras(args.load_path+'/'+args.prototxt, args.load_path+'/'+args.caffemodel, debug=args.debug)
  File "/home/ubuntu/keras/keras/caffe/convert.py", line 53, in caffe_to_keras
    debug)
  File "/home/ubuntu/keras/keras/caffe/convert.py", line 197, in create_model
    input_layers = ZeroPadding2D(padding=(pad_h, pad_w), name=name + '_zeropadding')(input_layers)
  File "/home/ubuntu/keras/keras/engine/topology.py", line 596, in __call__
    output = self.call(inputs, **kwargs)
  File "/home/ubuntu/keras/keras/layers/convolutional.py", line 1685, in call
    data_format=self.data_format)
  File "/home/ubuntu/keras/keras/backend/theano_backend.py", line 1124, in spatial_2d_padding
    input_shape = x.shape
AttributeError: 'list' object has no attribute 'shape'

[qxcv]

Not sure why that's happening. Are you sure you're actually using the code from this repo? My line 197 of models.py is different from the line 197 you have there, and the module name is also wrong. This code in this repo is based on the converter in @MarcBS Keras fork—based on module names, I suspect you're actually using that.