You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I used MinkownskiConvolution with a stride of 2 and kernel size 7. My input is a 2D image where every 3rd pixel is zeroed out. The input size is [224, 224] and sparsity looks something as follows:
4 4 0 6
4 4 0 6
0 0 0 0
6 6 0 9 ..
...
I am not sure what is happening here. In my understanding, if the kernel center is non-active element, no computation is performed, and zeros are fitted between elements/pixels. But here it seems that some of the elements (e.g. in first row of the returned matrix 4 4 0 6, second 4 is returned when 0 is centre in first row of original matrix i.e. [1 0 {0} 1] ). Kindly let me know what is going on here.
To Reproduce
a = torch.zeros((224,224))
a = torch.unsqueeze(a,0).unsqueeze(0).to(0)
a[:, :,::3,::3] = 1
with torch.no_grad():
me_dwconv.kernel[:] = torch.ones((49)).unsqueeze(1).unsqueeze(2).to(0)
# forward analytic test
output = me_dwconv(input)
out = output.dense()[0]
print(out[0,0,0:4,0:4])
Expected behavior
A clear and concise description of what you expected to happen.
4 0 0 6
0 0 0 0
0 0 0 0
6 0 0 9
Python version: [e.g. 3.8.18]
==========System==========
Linux-6.2.0-39-generic-x86_64-with-glibc2.17
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=22.04
DISTRIB_CODENAME=jammy
DISTRIB_DESCRIPTION="Ubuntu 22.04.2 LTS"
3.8.18 (default, Sep 11 2023, 13:40:15)
[GCC 11.2.0]
==========Pytorch==========
1.8.0+cu111
torch.cuda.is_available(): True
==========NVIDIA-SMI==========
/usr/bin/nvidia-smi
Driver Version 535.129.03
CUDA Version 12.2
VBIOS Version 95.02.18.80.55
Image Version G002.0000.00.03
GSP Firmware Version N/A
==========NVCC==========
/usr/bin/nvcc
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0
==========CC==========
/usr/bin/c++
c++ (Ubuntu 10.5.0-1ubuntu1~22.04) 10.5.0
Copyright (C) 2020 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
==========MinkowskiEngine==========
0.5.4
MinkowskiEngine compiled with CUDA Support: True
NVCC version MinkowskiEngine is compiled: 11050
CUDART version MinkowskiEngine is compiled: 11050
Additional context
Add any other context about the problem here.
The text was updated successfully, but these errors were encountered:
I used MinkownskiConvolution with a stride of 2 and kernel size 7. My input is a 2D image where every 3rd pixel is zeroed out. The input size is [224, 224] and sparsity looks something as follows:
1 0 0 1 0 0 1 0 0 1
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 0 0 1 0 0 1 0 0 1
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 0 0 1 0 0 1 0 0 1
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 0 0 1 0 0 1 0 0 1
.
.
I was expecting the output to be:
4 0 0 6
0 0 0 0
0 0 0 0
6 0 0 9 ..
...
But the output is :
4 4 0 6
4 4 0 6
0 0 0 0
6 6 0 9 ..
...
I am not sure what is happening here. In my understanding, if the kernel center is non-active element, no computation is performed, and zeros are fitted between elements/pixels. But here it seems that some of the elements (e.g. in first row of the returned matrix 4 4 0 6, second 4 is returned when 0 is centre in first row of original matrix i.e. [1 0 {0} 1] ). Kindly let me know what is going on here.
To Reproduce
a = torch.zeros((224,224))
a = torch.unsqueeze(a,0).unsqueeze(0).to(0)
a[:, :,::3,::3] = 1
input = to_sparse(a)
me_dwconv = MinkowskiConvolution(
in_channels, out_channels, kernel_size=7, stride=2, bias=False, dimension=D
).to(0)
with torch.no_grad():
me_dwconv.kernel[:] = torch.ones((49)).unsqueeze(1).unsqueeze(2).to(0)
# forward analytic test
output = me_dwconv(input)
out = output.dense()[0]
print(out[0,0,0:4,0:4])
Expected behavior
A clear and concise description of what you expected to happen.
4 0 0 6
0 0 0 0
0 0 0 0
6 0 0 9
==========System==========
Linux-6.2.0-39-generic-x86_64-with-glibc2.17
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=22.04
DISTRIB_CODENAME=jammy
DISTRIB_DESCRIPTION="Ubuntu 22.04.2 LTS"
3.8.18 (default, Sep 11 2023, 13:40:15)
[GCC 11.2.0]
==========Pytorch==========
1.8.0+cu111
torch.cuda.is_available(): True
==========NVIDIA-SMI==========
/usr/bin/nvidia-smi
Driver Version 535.129.03
CUDA Version 12.2
VBIOS Version 95.02.18.80.55
Image Version G002.0000.00.03
GSP Firmware Version N/A
==========NVCC==========
/usr/bin/nvcc
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0
==========CC==========
/usr/bin/c++
c++ (Ubuntu 10.5.0-1ubuntu1~22.04) 10.5.0
Copyright (C) 2020 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
==========MinkowskiEngine==========
0.5.4
MinkowskiEngine compiled with CUDA Support: True
NVCC version MinkowskiEngine is compiled: 11050
CUDART version MinkowskiEngine is compiled: 11050
Additional context
Add any other context about the problem here.
The text was updated successfully, but these errors were encountered: