Dataset Provenance: Built from
Master_H2.csv (423,515 reviewed images). Token and view are extracted from Filename: token is the segment preceding the 8-digit date, view is the segment following the UUID. Trailing shot indexes are consolidated (e.g., rear-right1 → rear-right).Short Form Coverage: 6,524 filenames use a short form with no
comparison- prefix (predominantly _stitched composites); they carry no view and infer their namespace from the leading filename segment.Exclusions & Multiplicity: Labels match case-insensitively.
Full Retouch and images with zero observations are excluded, leaving 419,706 flagged images. The Total row sums occurrences; an image with 3 logged observations contributes once to each issue's total.