Validate re-labeling of heterogeneous aircraft classes

Establish, through a better-powered experiment, whether re-labeling a heterogeneous aircraft class such as A330 into its constituent sub-variants improves detection performance beyond the effects of measurement noise, limited resolution, and minority-subclass scarcity.

Background

The framework attributes A330’s excess body-length dispersion to within-class heterogeneity arising from the A330-200 and A330-300 sub-variants. It therefore proposes re-labeling the class into sub-variants rather than adding more examples of the same combined label.

An initial threshold-based split produced essentially no performance improvement, and the paper explains that the test was underpowered and confounded by the limited geometric resolution and scarcity of the minority sub-class. Consequently, the proposed re-labeling remedy remains an unconfirmed prediction on this dataset.

References

We therefore state re-labeling as the distinctive fourth verdict, one no other quadrant produces, but as a prediction not yet confirmed on this dataset.

— Not All Confusion Is Equal: A Source-Aware Uncertainty Diagnosis for Fine-Grained Aircraft Detection  (2609.29959 - Huang et al., 24 Sep 2026) in Section 4.4, subsection “A330: heterogeneity, irreducible for a different reason”