[Glean] 3D Convolution: kernel will traverse in 3-D

Published: by Creative Commons Licence (Last updated: )

3D Convolution: kernel will traverse in 3-D

A 3D Convolution can be used to find patterns across 3 spatial dimensions; i.e. depth, height and width. One effective use of 3D Convolutions is object segmentation in 3D medical imaging. Since a 3D model is constructed from the medical image slices, this is a natural fit. And action detection in video is another popular research area, where multiple image frames are concatenated across a temporal dimension to give a 3D spatial input, and patterns are found across frames too.

Spatial dimensions are distinct from the channels dimension. Although color images have 3 channels, there are still only 2 spatial dimensions (of height and width), making 2D Convolutions more appropriate in this case. And RGB videos will still have 3 spatial dimensions (of time, height and width), making 3D Convolutions ideal.