Abstract
Out-of-Distribution (OoD) segmentation is critical for safety-sensitive applications like autonomous driving. However, existing mask-based methods often suffer from boundary imprecision, inconsistent anomaly scores within objects, and false positives from background noise. We propose \textbf{\textit{Objectomaly}}, an objectness-aware refinement framework that incorporates object-level priors. Objectomaly consists of three stages: (1) Coarse Anomaly Scoring (CAS) using an existing OoD backbone, (2) Objectness-Aware Score Calibration (OASC) leveraging SAM-generated instance masks for object-level score normalization, and (3) Meticulous Boundary Precision (MBP) applying Laplacian filtering and Gaussian smoothing for contour refinement. Objectomaly achieves state-of-the-art performance on key OoD segmentation benchmarks, including SMIYC AnomalyTrack/ObstacleTrack and RoadAnomaly, improving both pixel-level (AuPRC up to 96.99, FPR$_{95}$ down to 0.07) and component-level (F1$-$score up to 83.44) metrics. Ablation studies and qualitative results on real-world driving videos further validate the robustness and generalizability of our method. Code will be released upon publication.
Abstract (translated)
**翻译** 分布外（OoD，Out-of-Distribution）分割对于自动驾驶等安全敏感型应用至关重要。然而，现有的基于掩码的方法常常会遇到边界不精确、对象内部异常分数不一致以及背景噪声导致的假阳性的问题。为此，我们提出了**Objectomaly**——一个基于目标特性的改进框架，该框架利用了物体级别的先验知识。Objectomaly包括三个阶段：（1）粗略异常评分（CAS），使用现有的OoD骨干网络；（2）目标感知得分校准（OASC），通过SAM生成的实例掩码对对象级别的分数进行归一化处理；以及（3）细致边界精度（MBP），应用拉普拉斯滤波和高斯平滑以优化轮廓细节。在关键的OoD分割基准测试，如SMIYC AnomalyTrack/ObstacleTrack 和RoadAnomaly 上，Objectomaly实现了最先进的性能，在像素级别指标上取得了高达96.99的AuPRC值并把FPR$_{95}$降低到了0.07，并且在组件级别的F1-分数上达到了83.44。通过对实际驾驶视频中的消融研究和定性分析，进一步验证了我们方法的鲁棒性和泛化能力。代码将在发布后公开。
URL
https://arxiv.org/abs/2507.07460
