chainer.functions.upsampling_2d

chainer.functions.upsampling_2d(x, indexes, ksize, stride=None, pad=0, outsize=None, cover_all=True)[source]

Upsampling using pooling indices.

This function produces an upsampled image using pooling indices.

Example

It should be noted that you need to turn off chainer.config.use_cudnn flag when you perform max_pooling_2d() function which will make a pooling indicies for this upsampling_2d(). It is because indexes is never created and stored in the MaxPooling2D object when cuDNN is used for it.

>>> x = np.arange(1, 37).reshape(1, 1, 6, 6).astype('f')
>>> x = chainer.Variable(x)
>>> x.data
array([[[[  1.,   2.,   3.,   4.,   5.,   6.],
         [  7.,   8.,   9.,  10.,  11.,  12.],
         [ 13.,  14.,  15.,  16.,  17.,  18.],
         [ 19.,  20.,  21.,  22.,  23.,  24.],
         [ 25.,  26.,  27.,  28.,  29.,  30.],
         [ 31.,  32.,  33.,  34.,  35.,  36.]]]], dtype=float32)

This is the original x before max pooling.

>>> p = F.MaxPooling2D(2, 2)
>>> with chainer.using_config('use_cudnn', 'never'):
...     pooled_x = p.apply((x,))[0]
>>> pooled_x.data
array([[[[  8.,  10.,  12.],
         [ 20.,  22.,  24.],
         [ 32.,  34.,  36.]]]], dtype=float32)

This is the output of the max pooling operation. upsampling_2d() needs indexes array stored in the max pooling object p.

>>> upsampled_x = F.upsampling_2d(
...     pooled_x, p.indexes, p.kh, p.sy, p.ph, x.shape[2:])
>>> upsampled_x.shape
(1, 1, 6, 6)
>>> upsampled_x.data
array([[[[  0.,   0.,   0.,   0.,   0.,   0.],
         [  0.,   8.,   0.,  10.,   0.,  12.],
         [  0.,   0.,   0.,   0.,   0.,   0.],
         [  0.,  20.,   0.,  22.,   0.,  24.],
         [  0.,   0.,   0.,   0.,   0.,   0.],
         [  0.,  32.,   0.,  34.,   0.,  36.]]]], dtype=float32)
Parameters:
  • x (Variable) – Input variable.
  • indexes (ndarray or ndarray) – Index array that was used to calculate x with MaxPooling2D.
  • ksize (int or (int, int)) – ksize attribute of MaxPooling2D object that is used to calculate x
  • stride (int or (int, int)) – stride attribute of MaxPooling2D object that is used to calculate x
  • pad (int or (int, int)) – pad attribute of MaxPooling2D object that is used to calculate x
  • outsize ((int, int)) – Expected output size (height, width).
  • cover_all (bool) – Whether cover_all is used in the MaxPooling2D object or not.
Returns:

Output variable.
Return type:

Variable