English

Distilled Large Language Model-Driven Dynamic Sparse Expert Activation Mechanism

Computer Vision and Pattern Recognition 2026-03-31 v1 Artificial Intelligence

Abstract

High inter-class similarity, extreme scale variation, and limited computational budgets hinder reliable visual recognition across diverse real-world data. Existing vision-centric and cross-modal approaches often rely on rigid fusion mechanisms and heavy annotation pipelines, leading to sub-optimal generalization. We propose the Distilled Large Language Model (LLM)-Driven Sparse Mixture-of-Experts (DS-MoE) framework, which integrates text-guided dynamic routing and lightweight multi-scale comprehension. The DS-MoE framework dynamically aligns textual semantics with defect-specific visual patterns through a sparse MoE architecture, where task-relevant experts are adaptively activated based on semantic relevance, resolving inter-class ambiguity. A lightweight MobileSAM encoder enables real-time inference while preserving multi-scale defect details. Extensive experiments on PCB, aluminum foil, and mold defect datasets demonstrate that our framework achieves superior performance compared to existing pure vision models. \textbf{DS-MoE} surpasses YOLOv8/YOLOX with gains of +13.9, +1.4, and +2.0 pp mAP@ 0.5:0.95 on BBMP, aluminum, and PCB, respectively, while also improving precision and recall.

Keywords

Cite

@article{arxiv.2603.26735,
  title  = {Distilled Large Language Model-Driven Dynamic Sparse Expert Activation Mechanism},
  author = {Qinghui Chen and Zekai Zhang and Zaigui Zhang and Kai Zhang and Dagang Li and Wenmin Wang and Jinglin Zhang and Cong Liu},
  journal= {arXiv preprint arXiv:2603.26735},
  year   = {2026}
}
R2 v1 2026-07-01T11:41:24.197Z