I’m wondering how i can implement convolution layer with sparse 2D image data.
The example is attached.
It has patterns in a very little portion of the entire image and
the rest pixels of the image is zero.
These kinds of images are stacked in multiple channels.
Would the dilated convolution or submanifold convolution be helpful?
Or, Can i just use normal 2D convoluton layer for this data?
Please help me!
Thank you very much in advance.