Closed astrocyted closed 1 year ago
Hi, yes this was what I originally did but for various reasons the pytorch deformconv_2d
is not very good for interpretability and analysis or for example if you what control the minimum/maximum receptive field of the layer. This may not be important to many. Also if this is the kind of implementation you want I recommend another project called tvdcn
at github.com/inspiros/tvdcn
Hi,
wouldn't it have been way easier to write the deform_conv1d as a small wrapper around deform_conv2d() in which you "unsqueeze()" the 1D input (and kernal dims) and pass zeros() for the offset channel corresponding to the unsqueezed (fabricated) dimension ?