GenSeg is a supervised machine learning model based on SegNet that semantically segments n-dimensional data.
It can operate on data with N spatial dimensions instead of just two like in the original SegNet. This enables more complex data, such as volumetric data, to be used as inputs. However, due to restrictions in the way that convolutions are implemented in TensorFlow, this implementation of GenSeg currently works only for 1<=N<=3 spatial dimensions. Hence, 3D data such as LiDAR is supported, as well as more conventional data such as images, but higher dimensional data such as 3D time series data will not be supported until TensorFlow adds support for 4D and higher convolutions. If/When they do, this architecture should automatically work.
The output of GenSeg is a set of class probabilities for each pixel/voxel/n-dimensional point in the image being segmented. This allows for entire scenes to be understood similarly to how humans might understand them.