🤖 AI Summary
This work addresses the significant performance degradation of existing sparse anchor calibration methods under real-world sensor anomalies—particularly multipath interference—where anchors are present but corrupted by erroneous values. The authors propose MRAC, a training-free, inference-time calibration framework that, for the first time, exposes a structural blind spot in the widely used VI-Depth method under such conditions. MRAC introduces a robust, parameter- and K-value-agnostic mechanism that leverages the base model’s internal consistency to identify reliable anchors, followed by Theil–Sen estimation combined with Median Absolute Deviation (MAD) testing. The entire calibration process completes in approximately 50 microseconds on a CPU and supports arbitrary numbers of anchors. Evaluated on a 320-instance benchmark, MRAC achieves an 84% win rate and reduces AbsRel error by 3.2× (from 0.489 to 0.151) on KITTI multipath scenarios—all without requiring model retraining.
📝 Abstract
Monocular depth foundations predict domain-general relative depth but lack absolute scale; a handful of sparse metric anchors from a range sensor can calibrate them to metric depth, an attractive alternative to metric-supervised training. Existing sparse-anchor calibration methods, however, assume the anchors are clean, whereas real sensors produce outliers that are present with the wrong value -- time-of-flight multipath, mixed pixels -- not merely missing. We show that the established residual-on-CFA calibration recipe collapses under such outliers, and that the strongest publicly deployed method, VI-Depth, has a structural multipath blind spot: robust to missing anchors, it falls behind an unprotected baseline on three of four datasets when anchors are present but wrong. We propose Multipath-Robust Anchor Calibration (MRAC), a parameter-free, inference-time wrapper that gates anchors by foundation consistency -- a Theil--Sen fit and a median-absolute-deviation test against the foundation's own relative-depth ordering -- before a single call to the calibration head. MRAC adds no learned parameters, runs its selection in $\approx 50\,μ$s on CPU, and serves anchor budgets $K \in [5,200]$ from one checkpoint. On a $320$-cell benchmark with a same-backbone, same-architecture control, MRAC strictly wins $84\%$ of same-backbone cells across all four outlier families and, against VI-Depth, wins all twelve corrupted multipath cells and all sixteen KITTI cells, reducing KITTI multipath AbsRel by $3.2\times$ ($0.489$ to $0.151$) at zero retraining.