Reliability of confidence-gated shadow-direction estimation

Determine whether a more selective confidence-gated strategy for ShadowCast direction estimation can improve the quality of synthetic training samples without reducing the segmentation benefit of shadow augmentation.

Background

ShadowCast estimates a single cast-shadow direction by selecting the darkest valid probe region around label-derived elevated-object masks. The supplementary experiments introduce two confidence checks—a brightness-gap test and an opposite-side brightness test—to reject ambiguous direction estimates.

The confidence-gated variant improves over the baseline but performs worse than the default estimator on all three benchmarks. The authors therefore identify an unresolved methodological question: whether direction-estimation reliability can be assessed or improved independently of the reduction in augmentation frequency, so that unreliable samples are removed without sacrificing the benefits of shadow augmentation.

References

Further work should evaluate direction reliability separately from augmentation frequency to determine whether a more selective strategy can improve sample quality without losing the segmentation benefit.

— ShadowCast: On-the-Fly Context-Aware Shadow Augmentation for Remote Sensing Image Segmentation  (2610.23440 - Kim et al., 16 Oct 2026) in Appendix, Section “Design Choices and Additional Ablation Studies of ShadowCast,” subsection “Confidence-Gated Direction Estimation”