Myriad: a large multimodal model applying vision experts for industrial anomaly detection

Industrial anomaly detection (IAD) has developed strong methods for one-class, zero-shot, and few-shot deployments, each effective in its own setting, yet no single approach adapts quickly across them. Large multimodal models (LMMs) offer a different route, since broad knowledge and strong visual-language understanding allow an LMM to interpret the outputs of diverse IAD methods and to calibrate their predictions using image evidence. To realize such an adaptive system, we present a novel large multimodal model applying vision experts for industrial anomaly detection (abbreviated as Myriad). Myriad treats conventional IAD models as VEs and converts their anomaly maps into lightweight prompts that steer a frozen Q-Former toward suspicious regions, while a compact low-rank adapter shapes features for IAD. The language pathway then fuses VE positional cues with visual evidence to produce calibrated, machine-usable decisions, effectively regularizing noisy or ambiguous anomaly maps. By simply switching the VE (e.g., one-class or zero-/few-shot experts) without modifying the architecture, Myriad adapts uniformly across deployment scenarios and remains robust to the choice of expert. Extensive experiments on MVTec-AD, VisA, and PCB Bank benchmarks demonstrate that our proposed method not only performs favorably against state-of-the-art methods under one-class and few-shot settings, but also inherits the flexibility and instruction-following ability of LMMs in the field of IAD. Source code and pre-trained models are publicly available at https://github.com/tzjtatata/Myriad.