Labeling components in a binary tensor

vision
1

Hi Everyone, I have a binary image which is the output of a segmentation net. As part of the post processing I have to find all the connected components in this image, which I do with the openCV connectedComponents function. However I was wondering if there is a pytorch equivalent so I can do this operation while the tensor is still on the GPU. Does anyone have or know of something equivalent or similar?

Thanks,
Dave

1 Like
2

Hi David!

I don’t know one way or the other whether there is a pytorch / gpu
connected-components function. (I sort of doubt it.) But, in any
event, a connected-components algorithm would be unlikely to
benefit much from a gpu’s pipelined architecture.

I suppose that if after constructing the connected components
you needed to move the result back to the gpu, avoiding the
gpu-cpu-gpu memory transfers could be a benefit of running
connected components on the gpu.

Best.

K. Frank

3

Hi K, thanks for your reply. I haven’t yet found anything similar in pytorch so perhaps you’re right. Will keep looking though.

Dave

4

A GPU connected component implementation https://developer.download.nvidia.com/video/gputechconf/gtc/2019/presentation/s9111-a-new-direct-connected-component-labeling-and-analysis-algorithm-for-gpus.pdf

I think this can easily be taken in.