Video anomaly detection (VAD) faces significant challenges in multimodal semantic alignment and long-term temporal modeling within open surveillance scenarios. Existing methods are often plagued by modality discrepancies and fragmented temporal reasoning. To address these issues, we introduce MT-CMVAD, a hierarchically structured Transformer architecture that makes two key technical contributions: (1) A Context-Aware Dynamic Fusion Module that leverages cross-modal attention with learnable gating coefficients to effectively bridge the gap between RGB and optical flow modalities through adaptive feature recalibration, significantly enhancing fusion performance; (2) A Multi-Scale Spatiotemporal Transformer that establishes global-temporal dependencies via dilated attention mechanisms while preserving local spatial semantics through pyramidal feature aggregation. To address the sparse anomaly supervision dilemma, we propose a hybrid learning objective that integrates dual-stream reconstruction loss with prototype-based contrastive discrimination, enabling the joint optimization of pattern restoration and discriminative representation learning. Our extensive experiments on the UCF-Crime, UBI-Fights, and UBnormal datasets demonstrate state-of-the-art performance, achieving AUC scores of 98.9%, 94.7%, and 82.9%, respectively. The explicit spatiotemporal encoding scheme further improves temporal alignment accuracy by 2.4%, contributing to enhanced anomaly localization and overall detection accuracy. Additionally, the proposed framework achieves a 14.3% reduction in FLOPs and demonstrates 18.7% faster convergence during training, highlighting its practical value for real-world deployment. Our optimized window-shift attention mechanism also reduces computational complexity, making MT-CMVAD a robust and efficient solution for safety-critical video understanding tasks.

Lightweight visual models are crucial for industrial defect detection tasks. Traditional methods and even some lightweight detectors often struggle with the trade-off between high computational demands and insufficient accuracy. To overcome these issues, this study introduces YOLO-MAD, an innovative model optimized through a multi-scale geometric structure feature extraction and fusion scheme. YOLO-MAD integrates three key modules: AKConv for robust geometric feature extraction, BiFPN to facilitate effective multi-scale feature integration, and Detect_DyHead for dynamic optimization of detection capabilities. Empirical evaluations demonstrate significant performance improvements: YOLO-MAD achieves a 5.4% mAP increase on the NEU-DET dataset and a 4.8% mAP increase on the GC10-DET dataset. Crucially, this is achieved under a moderate computational load (9.4 GFLOPs), outperforming several prominent lightweight models in detection accuracy while maintaining comparable efficiency. The model also shows enhanced recognition performance for most defect categories. This work presents a pioneering approach that balances lightweight design with high detection performance by efficiently leveraging multi-scale geometric feature extraction and fusion, offering a new paradigm for industrial defect detection.