CoGeoAD: Hierarchical Color-Geometric Fusion with Multi-View Attention for Zero-Shot 3D Anomaly Detection

Zero-shot 3D anomaly detection is essential for industrial quality inspection, where labeled anomaly samples are scarce. Meanwhile, existing methods lack an effective mechanism to fuse complementary 2D color images with 3D geometric structures, limiting their ability to detect both surface and structural defects in a unified framework. To address these issues, we propose CoGeoAD, a unified CLIP-based framework that fuses color and geometric features by constructing pixel-aligned paired multi-view images. The framework introduces a Data-Driven Multi-View Attention (MVA) mechanism to adaptively aggregate 3D features and a Multi-Stage Color-Geometric Fusion (MS-CGF) module to hierarchically integrate multi-level features from both modalities. Extensive experiments on the MVTec3D-AD and Eyecandies benchmarks demonstrate that CoGeoAD achieves state-of-the-art performance, effectively capturing both structural and textural anomalies in complex industrial scenarios.

In manufacturing, catching defective products is crucial, but it is highly difficult to train computers to do this when examples of real defects are scarce. Most current systems for spotting these flaws rely only on regular color photos, which can easily miss 3D structural issues like dents or cracks, whereas systems that only look at 3D shapes often miss color defects like surface stains. To solve this, we developed CoGeoAD, a new zero-shot system that effectively combines both color and 3D shape information without needing to be trained on specific examples of defects beforehand. Our system takes the 3D object and generates multiple 2D pictures from various angles, explicitly separating the object's physical shape and its true color. It then uses a smart, data-driven "attention" mechanism that automatically figures out which viewpoints are the most important for spotting a potential flaw. By carefully hierarchically merging this color and shape information, the system can accurately find both structural damage and surface blemishes simultaneously. When tested on complex industrial datasets, CoGeoAD significantly outperformed previous methods, proving to be a highly effective and reliable tool for automated quality control.