YOLO-ULM：面向实时目标检测的超轻量模型
YOLO-ULM: Ultra-Lightweight Models for Real-Time Object Detection
提供了一套可复用的高效检测器设计范式：并行下采样与高效特征聚合模块可在不损失信息的前提下压缩骨干网络，配合改进的完整 IoU 损失，可直接迁移到其他实时检测或轻量骨干网络设计中。
原文依据 YOLO-ULM-N achieves 41.6% mAP with an inference latency of 1.52 ms
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论YOLO 系列以优异的精度与速度引领目标检测。然而，基于卷积和自注意力的架构都存在参数冗余与计算效率不足的问题。现有轻量化方法过度追求速度，却忽视了特征提取以及跨不同阶段的空间变换过程中重要信息的损失。因此，有效的轻量化对检测性能至关重要。我们提出 YOLO-ULM，一种超轻量实时检测器，在保持高精度的同时实现推理加速。我们创新性地设计了多种由效率与精度双重驱动的模块，包括高效特征聚合模块与并行下采样模块，以及一个更聚焦的完整 IoU 损失函数。为验证所提方法，我们在 COCO 数据集上从零开始训练，不使用预训练权重。通过精简骨干网络参数，我们将其扩展为 YOLO-ULM-Turbo 以实现推理加速。YOLO-ULM 超越了 YOLOv11/YOLOv12/YOLOv13 和 RT-DETR 等最优实时检测器。在 T4 GPU 上，YOLO-ULM-N 取得 41.6% 的平均精度均值，推理延迟为 1.52 ms，优于 YOLOv11-N（提升 2.2%）和 YOLOv12-N（提升 1.0%）。YOLO-ULM-S 以少 64.7% 的浮点运算量和少 63% 的参数超越 RT-DETR-R18 达 1.6% 的平均精度均值。YOLO-ULM-L / X 的平均精度均值分别比 YOLOv13-L / X 高 0.7% 和 0.8%。YOLO-ULM-Turbo 达到与 YOLOv12-Turbo 相当的性能，但计算量更少，其中 Turbo-N 变体相比 YOLOv12-Turbo-N 平均精度均值高 0.3%、参数少 16%。
英文原文摘要 1573 字符
YOLO series lead object detection with superior accuracy and speed. However, both convolutional and self-attention based architectures suffer from parameter redundancy and insufficient computational efficiency. Existing lightweight methods excessively pursue speed while ignoring the loss of important information during feature extraction and spatial transformation across different stages. Thus, effective lightweighting is crucial for detection performance. We propose YOLO-ULM, an ultra-lightweight real-time detector that achieves accelerated inference while preserving high accuracy. We innovatively design a variety of dual efficiency- and accuracy-driven modules, including efficient feature aggregation and parallel downsampling modules, as well as a more focused complete-IoU loss function. To validate our approach, we train it from scratch on COCO dataset without pretrained weights. By refining backbone parameters, we extend it to YOLO-ULM-Turbo for accelerated inference. YOLO-ULM surpasses state-of-the-art real-time detectors like YOLOv11/YOLOv12/YOLOv13 and RT-DETR. On a T4 GPU, YOLO-ULM-N achieves 41.6% mAP with an inference latency of 1.52 ms, outperforming YOLOv11-N (2.2%\uparrow) and YOLOv12-N (1.0%\uparrow). YOLO-ULM-S exceeds RT-DETR-R18 by 1.6% mAP with 64.7% fewer FLOPs and 63% fewer parameters. YOLO-ULM-L / X surpass YOLOv13-L / X by 0.7% and 0.8% respectively in mAP. YOLO-ULM-Turbo matches YOLOv12-Turbo's performance but uses less computation, with Turbo-N variant achieving 0.3% higher mAP and 16% fewer parameters than YOLOv12-Turbo-N.