A Boundary-Metric Evaluation Protocol for Whiteboard Stroke Segmentation Under Extreme Imbalance

TL;DR

This paper introduces a comprehensive evaluation protocol for whiteboard stroke segmentation that accounts for extreme class imbalance and thin-stroke accuracy, revealing trade-offs between mean performance and worst-case reliability.

Contribution

It proposes a novel evaluation protocol combining region, boundary, and equity metrics, and demonstrates how different loss functions and resolutions affect segmentation performance and robustness.

Findings

Overlap-based losses significantly improve F1 scores.

Boundary metrics confirm contour accuracy improvements.

Higher resolution training boosts F1 by 12.7 points.

Abstract

The binary segmentation of whiteboard strokes is hindered by extreme class imbalance, caused by stroke pixels that constitute only $1.79%$ of the image on average, and in addition, the thin-stroke subset averages $1.14% \pm 0.41%$ in the foreground. Standard region metrics (F1, IoU) can mask thin-stroke failures because the vast majority of the background dominates the score. In contrast, adding boundary-aware metrics and a thin-subset equity analysis changes how loss functions rank and exposes hidden trade-offs. We contribute an evaluation protocol that jointly examines region metrics, boundary metrics (BF1, B-IoU), a core/thin-subset equity analysis, and per-image robustness statistics (median, IQR, worst-case) under seeded, multi-run training with non-parametric significance testing. Five losses -- cross-entropy, focal, Dice, Dice+focal, and Tversky -- are trained three times each on a DeepLabV3-MobileNetV3 model and evaluated on 12 held-out images split into core and thin subsets. Overlap-based losses improve F1 by more than 20 points over cross-entropy ($0.663$ vs $0.438$, $p < 0.001$). In addition, the boundary metrics confirm that the gain extends to the precision of the contour. Adaptive thresholding and Sauvola binarization at native resolution achieve a higher mean F1 ($0.787$ for Sauvola) but with substantially worse worst-case performance (F1 $= 0.452$ vs $0.565$ for Tversky), exposing a consistency-accuracy trade-off: classical baselines lead on mean F1 while the learned model delivers higher worst-case reliability. Doubling training resolution further increases F1 by 12.7 points.