Tracker sometimes crashes when running addψs!(props::Vector{DataFrame})

[hollandjg]

Reproduce using props object: https://drive.google.com/open?id=191DvguUWgWJYM681YJziPAZ4QWgWfThJ&usp=drive_fs

[cpaniaguam]

It's an issue with the input data with blobs whose boundaries can't be resampled. The workflow includes routines to trace the boundaries of blobs whose boundaries are closed (start point and end point match) for shape recognition. A blob that looks like a line segment fails to have this property.

julia> props = deserialize("tracker-bug/psi-s-bug/props.jls");

julia> props[2][870, :area]
2.0

julia> props[2][870, :mask]
3×3 BitMatrix:
 0  0  0
 0  1  0
 0  0  1

Elsewhere I suggested filtering out blobs that are too small. The old workflow used a default range (350, 90_000) for blob areas.

[hollandjg]

I'm not sure this is fixed. Can we reopen? A blob which is one pixel in size isn't pointlike, as you could treat its corners as its boundary.