Odd-One-Out: Anomaly Detection by Comparing with Neighbors
This addresses a novel anomaly detection task for scene-specific anomalies, which is incremental as it builds on existing AD methods by adding multi-view and 3D modeling components.
The paper tackles the problem of detecting anomalous objects within a scene by comparing them to other objects, using multiple views to build 3D object-centric models and part-aware representations for cross-instance comparison, and introduces two new benchmarks (ToysAD-8K and PartsAD-15K) for evaluation.
This paper introduces a novel anomaly detection (AD) problem aimed at identifying `odd-looking' objects within a scene by comparing them to other objects present. Unlike traditional AD benchmarks with fixed anomaly criteria, our task detects anomalies specific to each scene by inferring a reference group of regular objects. To address occlusions, we use multiple views of each scene as input, construct 3D object-centric models for each instance from 2D views, enhancing these models with geometrically consistent part-aware representations. Anomalous objects are then detected through cross-instance comparison. We also introduce two new benchmarks, ToysAD-8K and PartsAD-15K as testbeds for future research in this task. We provide a comprehensive analysis of our method quantitatively and qualitatively on these benchmarks.