diff options
author | Michael 'myrhev' Mathieu <michael.mathieu@ens.fr> | 2012-03-24 05:18:05 +0400 |
---|---|---|
committer | Michael 'myrhev' Mathieu <michael.mathieu@ens.fr> | 2012-03-24 05:20:19 +0400 |
commit | 12ba058c41397436cd3c80c14f7375cb77e81cbb (patch) | |
tree | 19bfd08313fa81d0948ac85fc3f889508052e734 /SpatialUpSampling.lua | |
parent | 111dc87bd431717e3a71425b9b34c70f778d0f5b (diff) |
Fix bug in SpatialUpSampling whith non-contiguous tensors
Diffstat (limited to 'SpatialUpSampling.lua')
-rw-r--r-- | SpatialUpSampling.lua | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/SpatialUpSampling.lua b/SpatialUpSampling.lua index 0d26bcd..43b9de6 100644 --- a/SpatialUpSampling.lua +++ b/SpatialUpSampling.lua @@ -3,15 +3,15 @@ local SpatialUpSampling, parent = torch.class('nn.SpatialUpSampling', 'nn.Module local help_desc = [[ Applies a 2D up-sampling over an input image composed of several input planes. The input tensor in forward(input) is -expected to be a 3D tensor (width x height x nInputPlane). +expected to be a 3D tensor (nInputPlane x width x height). The number of output planes will be the same as nInputPlane. The upsampling is done using the simple nearest neighbor technique. For interpolated (bicubic) upsampling, use nn.SpatialReSampling(). -If the input image is a 3D tensor width x height x nInputPlane, -the output image size will be owidth x oheight x nInputPlane where +If the input image is a 3D tensor nInputPlane x width x height, +the output image size will be nInputPlane x owidth x oheight where owidth = width*dW oheight = height*dH ]] |