I am working on an object segmentation problem in AR space and we are using ARKit2. ARKit has this object scanning and detection capability which finds out interesting points (features) and uses them for detection.
I was thinking of clustering these extracted features and segmenting the object using a weakly-supervised approach. ARKit gives access to these points through
ARWorldMap.rawFeaturePoints.points
but these are only spatial features (Swift array of float3 vectors - each 3D vector is a representation of one feature point in 3D AR space).Obviously ARKit is generating dense features based on edges, corners, texture & other things to get these points but it only exposes the 3D coordinates of the points instead of actual features. Is there any way to get those "real" features?