Identity mapping causes zero rows

[daviddao]

Using an identity transformation in transfromerlayer_test.py with all weights set to zero :

Expected output_train shape (1, 1200, 1600, 3)
Actual output_train shape (1, 1200, 1600, 3)
Transformation matrix:
[[ 1.  0.  0.  0.  1.  0.]]

I get following image matrix:

However the original image is:

Does the row full of zeros have sth to do with the bilinear interpolation? Is it supposed to be like this? Here the case with 100 fold downsampling:

[elezar]

Is this perhaps related to https://github.com/tensorflow/models/issues/193#issuecomment-225634013?

There it is stated that in the interpolation, the mapping should be between [-1, 1] and [0, width-1]. It seems as if the mapping has been corrected in https://github.com/Lasagne/Lasagne/commit/c8572b29331ffe03fa698eaae17351ab136a4a71, but not here.