Breaking the Horizontal Prior: From Long-Tailed Orientation Bias to Roll-Robust Monocular Depth Estimation
For practitioners deploying monocular depth models in real-world scenarios where camera orientation may vary, this work addresses a critical robustness gap, though the gains are specific to roll robustness rather than a broad improvement.
The paper identifies a long-tailed orientation bias (Horizontal Prior) in monocular depth estimation models, causing performance degradation under camera roll. They propose a training-time supervision strategy (ID-Constraint) that improves roll robustness by fine-tuning the backbone with auxiliary geometric tasks, achieving consistent improvements across five benchmarks and four roll settings.
Despite recent advances in Monocular Depth Estimation, state-of-the-art depth foundation models remain vulnerable to robustness issues. Particularly, even slight camera rolls can result in substantial degradation in depth estimations. We attribute this problem to a previously overlooked phenomenon, termed the Horizontal Prior, which is a manifestation of long-tailed distribution bias: most training images are captured in approximately horizontal orientations due to human visual preferences and photographic habits. While intuitive remedies such as re-balanced data augmentation and horizon leveling provide partial improvements, they fail to fully address the issue. In this paper, we introduce Invariant Depth Constraint (ID-Constraint), a training-time supervision strategy that improves roll robustness by fine-tuning and jointly regularizing the depth backbone with a series of geometric and spatial reasoning tasks. These auxiliary objectives encourage the backbone to learn rotation-stable, depth-relevant representations, while the auxiliary prediction heads are discarded after training, leaving the original inference architecture unchanged. Extensive experiments on five benchmark datasets across four roll settings demonstrate the effectiveness of the proposed method.