Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Creating the Coordinate Map directly #562

Open
HaFred opened this issue Sep 8, 2023 · 0 comments
Open

Creating the Coordinate Map directly #562

HaFred opened this issue Sep 8, 2023 · 0 comments

Comments

@HaFred
Copy link

HaFred commented Sep 8, 2023

Is your feature request related to a problem? Please describe.
I have an input sparse tensor to a sparse NN, looking like the below:

SparseTensor(
  coordinates=tensor([[  0, 308, 251, 100],
        [  0, 307, 252,  97],
        [  0, 307, 252,  99],
        ...,
        [  0,  83, 249,  90],
        [  0,  83, 249,  89],
        [  0,  83, 246,  97]], device='cuda:0', dtype=torch.int32)
  features=tensor([[ 0.0118,  0.0902,  0.4275],
        [ 0.1137,  0.0588,  0.4431],
        [ 0.1294,  0.2000,  0.4745],
        ...,
        [-0.4275, -0.3961,  0.6941],
        [-0.1137,  0.0353,  0.6549],
        [-0.2157, -0.1294,  0.2784]], device='cuda:0')
  coordinate_map_key=coordinate map key:[1, 1, 1]
  coordinate_manager=CoordinateMapManagerGPU_c10(
	[1, 1, 1, ��]:	CoordinateMapGPU:68270x4
	[2, 2, 2, ��]:	CoordinateMapGPU:33366x4
	[4, 4, 4, ��]:	CoordinateMapGPU:9637x4
	[8, 8, 8, ��]:	CoordinateMapGPU:2447x4
	[4, 4, 4, ��]->[8, 8, 8, ��]:	gpu_kernel_map: number of unique maps:8, kernel map size:9637
	[8, 8, 8, ��]->[8, 8, 8, ��]:	gpu_kernel_map: number of unique maps:27, kernel map size:38253
	[4, 4, 4, ��]->[4, 4, 4, ��]:	gpu_kernel_map: number of unique maps:27, kernel map size:141005
	[2, 2, 2, ��]->[4, 4, 4, ��]:	gpu_kernel_map: number of unique maps:8, kernel map size:33366
	[2, 2, 2, ��]->[2, 2, 2, ��]:	gpu_kernel_map: number of unique maps:27, kernel map size:406254
	[1, 1, 1, ��]->[2, 2, 2, ��]:	gpu_kernel_map: number of unique maps:8, kernel map size:68270
	[1, 1, 1, ��]->[1, 1, 1, ��]:	gpu_kernel_map: number of unique maps:125, kernel map size:1380550
	algorithm=MinkowskiAlgorithm.DEFAULT
  )
  spatial dimension=3)

However, I would like to create such a sparse tensor without passing to a new NN directly, by assigning its coordinate map. It seems like there are no current solutions for this.

Describe alternatives you've considered
I tried below:

sparse_tensor = ME.SparseTensor(feat, coords, tensor_stride=4)

But it cannot return the multiple-strides sparse tensor like the snippets above.

Thanks in advance.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant