Spatial data

Before introducing various projective transformations, we have a look at the data that can be projected. There are three kinds of data currently supported: images, keypoints and segmentation masks. Both 2D and 3D data is supported (and technically, higher dimensions, but I’ve yet to find a dataset with 4 spatial dimensions).

• Image{N, T} represents an N-dimensional image. T refers to the element type of the array that Image wraps, usually a color. When projecting images, proper interpolation methods are used to reduce artifacts like aliasing. See image.jl

• MaskBinary{N} and MaskMulti{N, T} likewise represents N-dimensional segmentation masks. Unlike images, nearest-neighbor interpolation is used for projecting masks. See mask.jl

• Lastly, Keypoints{N} represent keypoint data. The data should be an array of SVector{N}. Since there are many interpretations of keypoint data, there are also wrapper items for convenience: BoundingBox and Polygon.See keypoints.jl