深度学习论文: YOLO-Master: MOE-Accelerated with Specialized Transformers for Enhanced Real-time Detection

深度学习论文: YOLO-Master: MOE-Accelerated with Specialized Transformers for Enhanced Real-time DetectionYOLO-Master: MOE-Accelerated with Specialized Transformers for Enhanced Real-time DetectionPDF: https://arxiv.org/pdf/2512.23273PyTorch代码: https://github.com/shanglianlm0525/CvPytorchPyTorch代码: https://github.com/shanglianlm0525/PyTorch-Networks1 概述现有实时目标检测（RTOD）方法多采用类YOLO架构，兼顾精度与速度，但这类模型采用静态密集计算，对所有输入统一处理，导致表征能力与算力分配失衡：简单场景算力冗余、复杂场景算力不足，既引发计算浪费，又拉低检测性能。