Reference data quality impacts building segmentation

Performance rankings vary significantly depending on reference labels, with manual annotations showing substantially greater improvement (0.128 ± 0.036 IoU gain) than hyperparameter tuning (0.008 ± 0.010). The results demonstrate that reference data quality critically influences model evaluation outcomes and should be treated as a key factor alongside architecture selection.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.