In Theano, how to do Reverse-MaxPooling in the Convolutional MaxPooling Auto-Encoder

MaxPooling Layer always follows Convolutional Layer. But in the Convolutional MaxPooling Auto-Encoder, how to do reverse or Un-MaxPooling?

Here is the method:

neighbor box:
(2, 2)
original images:
[[[[ 15.  46.  46.  37.]
[ 73.  93.   2.  91.]
[ 97.  93.  78.   7.]
[  0.  24.  34.  26.]]]

[[[ 70.  46.   2.  93.]
[ 11.  54.  29.  83.]
[  0.  48.  69.  95.]
[ 49.  48.  10.  19.]]]]
neighbor flatten:
[[ 15.  46.  73.  93.]
[ 46.  37.   2.  91.]
[ 97.  93.   0.  24.]
[ 78.   7.  34.  26.]
[ 70.  46.  11.  54.]
[  2.  93.  29.  83.]
[  0.  48.  49.  48.]
[ 69.  95.  10.  19.]]
non maximum suppression result:
[[[[  0.   0.   0.   0.]
[  0.  93.   0.  91.]
[ 97.   0.  78.   0.]
[  0.   0.   0.   0.]]]

[[[ 70.   0.   0.  93.]
[  0.   0.   0.   0.]
[  0.   0.   0.  95.]
[ 49.   0.   0.   0.]]]]

View Code