CharlesShang / DCNv2

Deformable Convolutional Networks v2 with Pytorch
BSD 3-Clause "New" or "Revised" License
1.28k stars 402 forks source link

3D Deformable Conv..... #4

Open JarveeLee opened 5 years ago

JarveeLee commented 5 years ago

Could you please write a 3D version of deformable convolution ... I am dealing with medical data which require 3D conv but my modified 3D deformable conv from your code can not pass gradient check if threshold is small.

JarveeLee commented 5 years ago

my main modification is the bp for d-dimension like here,now my offset grad error is around 1e-3... grad img , grad mask , grad bias has been lower than 1e-5.... image

JarveeLee commented 5 years ago

But offset grad error from your 2D implementation can reach 1e-5...I don't know what is wrong here...

jmarsil commented 4 years ago

@JarveeLee have you come up with an adequate 3D solution that could be shared?

Kindpire commented 4 years ago

But offset grad error from your 2D implementation can reach 1e-5...I don't know what is wrong here...

Hi @CharlesShang @Jarveelee, I was trying to do the same thing but stuck in function dmcn_im2col_bilinear in deform_im2col_cuda.cuh, any hint about it?

hua-wu-que commented 4 years ago

any update? Any one knows does the im2col implementation suitable for 3d conv? (will it be too big for memory)