AI RESEARCH
Synthesis4AD: Synthetic Anomalies are All You Need for 3D Anomaly Detection
arXiv CS.CV
•
ArXi:2604.04658v1 Announce Type: new Industrial 3D anomaly detection performance is fundamentally constrained by the scarcity and long-tailed distribution of abnormal samples. To address this challenge, we propose Synthesis4AD, an end-to-end paradigm that leverages large-scale, high-fidelity synthetic anomalies to learn discriminative representations for 3D anomaly detection.