Padding 2D /3D Variable

1

I am trying to do padding a 2D matrix, when I use ConstantPad2d, it works fine for 4D but it is not working for 3D/2D.

import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from torch import autograd



input = autograd.Variable(torch.randn(1,1,3, 3))

m = nn.ConstantPad2d((1,1,1,1), 0)
output = m(input)
print(output)

Can anyone help me here.
I tried the other methods in the forum but couldn’t make them work.

basically this does not work:

input = autograd.Variable(torch.randn(3, 3))

m = nn.ConstantPad2d((1,1,1,1), 0)
output = m(input)
print(output)
2

What exactly does your output look like?

I ran the code and it looks like it does what it’s supposed to:
image

image

1 Like
3

That’s weird:
I get this error:

---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
<ipython-input-8-764c673fd541> in <module>()
     10 
     11 m = nn.ConstantPad2d((1,1,1,1), 0)
---> 12 output = m(input)
     13 print(output)

/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.pyc in __call__(self, *input, **kwargs)
    222         for hook in self._forward_pre_hooks.values():
    223             hook(self, input)
--> 224         result = self.forward(*input, **kwargs)
    225         for hook in self._forward_hooks.values():
    226             hook_result = hook(self, input, result)

/usr/local/lib/python2.7/dist-packages/torch/nn/modules/padding.pyc in forward(self, input)
    178 
    179     def forward(self, input):
--> 180         return F.pad(input, self.padding, 'constant', self.value)
    181 
    182     def __repr__(self):

/usr/local/lib/python2.7/dist-packages/torch/nn/functional.pyc in pad(input, pad, mode, value)
   1043             return _functions.thnn.ReplicationPad3d.apply(input, *pad)
   1044     else:
-> 1045         raise NotImplementedError("Only 4D and 5D padding is supported for now")
   1046 
   1047 

NotImplementedError: Only 4D and 5D padding is supported for now

torch.__version__

gives me:

‘0.2.0_3’

1 Like
4

Looks like the functionality was recently implemented. You could build pytorch from source if you need it (instructions here), or wait for the next release.

5

Yeah I guess so, I hope they fix it on the new release!
bummer :confused: