Guoyang Xie

Forget painstakingly collecting real-world defect data: high-fidelity synthetic anomalies, automatically generated from product designs using an MLLM, can dramatically improve 3D anomaly detection.

Naively combining unimodal architectures for multimodal anomaly detection makes catastrophic forgetting *worse* – unless you disentangle features with Mamba and filter redundancies with an information bottleneck.