AnomalyAgent: Agentic Industrial Anomaly Synthesis via Tool-Augmented Reinforcement Learning
AnomalyAgent is an agent designed to create realistic industrial anomalies by using five specific tools and iterative self-reflection. It uses a two-stage training process—supervised fine-tuning followed by reinforcement learning—driven by a three-part reward system to improve its ability to generate diverse, high-quality anomaly samples, outperforming existing methods.