[hannalee2]

Problem

Cropped section images showed thin one-pixel lines ending in small blobs, and narrow bridges to neighbouring fragments. They showed up in the segmented output.

Cause

Results.masks.xy builds its polygons with ops.masks2segments(), and its default strategy is "all". When a mask has several disconnected pieces (the section plus a debris fragment or a neighbouring section), "all" joins every contour into one polygon with thin connecting lines (merge_multi_segment). fillPoly in post_processing.py then kept the connectors and stray pieces in the crop.

Checked against Ultralytics 8.4.54:

• masks2segments(masks, strategy="all") is the default: link
• merge_multi_segment joins contours "with a thin line": ultralytics/data/converter.py
• Report of the same issue: ultralytics#17876

Changes

• Build each section polygon from the raw binary mask (_largest_polygon), keeping only the largest contour by area, so connectors and stray pieces are dropped.
• Open the mask with a small kernel to remove spurs and thin bridges. The kernel size is the new yolo.mask_open_kernel setting (default 3, 0 disables), added to the YAML, the QC schema and the web schema.

Example)
(Before)

(After)