Age | Commit message (Collapse) | Author |
|
|
|
|
|
Exceptions are:
1) SparseLinear
requires additional parameters to be passed in (e.g. nbatches),
so it's not clear it's worth moving to C since it won't really simplify the binding
code logic.
2) BatchNormalization
requires "makeBatch", which isn't a trivial translation to C.
3) LookupTable
requires "view" in C, which is already a TODO
4) SpatialUpSamplingBilinear
requires "view" in C, which is already TODO
|
|
VolumetricFullConvolution which can be used to dynamically set the output size, as an alternative to using the adj terms, which has the downside to be fixed at construction time.
|
|
Ideally while displaying the kernel (and likewise for padding, etc.), it should be of the format kT x kW x kH, as in the doc or normal Volumetric Convolution.
Here it was of the format kW x kH x kT. This caused confusion to me while creating a model and thought it might to others as well.
|
|
|
|
SpatialFullConvolution.
|
|
|
|
|
|
|