[论文解读] DAMO-YOLO : A Report on Real-Time Object Detection Design

Xianzhe Xu, Yiqi Jiang+4 more|arXiv (Cornell University)|Nov 23, 2022

Advanced Neural Network Applications被引用 118

一句话总结

DAMO-YOLO 在 YOLO 的基础上扩展了 NAS-backed 骨干网、一个高效的 RepGFPN 颈部网络、一个紧凑的 ZeroHead、AlignOTA 标签分配，以及蒸馏，以在 COCO 数据集上实现针对通用和轻量模型的实时对象检测的最新性能。

ABSTRACT

In this report, we present a fast and accurate object detection method dubbed DAMO-YOLO, which achieves higher performance than the state-of-the-art YOLO series. DAMO-YOLO is extended from YOLO with some new technologies, including Neural Architecture Search (NAS), efficient Reparameterized Generalized-FPN (RepGFPN), a lightweight head with AlignedOTA label assignment, and distillation enhancement. In particular, we use MAE-NAS, a method guided by the principle of maximum entropy, to search our detection backbone under the constraints of low latency and high performance, producing ResNet/CSP-like structures with spatial pyramid pooling and focus modules. In the design of necks and heads, we follow the rule of ``large neck, small head''.We import Generalized-FPN with accelerated queen-fusion to build the detector neck and upgrade its CSPNet with efficient layer aggregation networks (ELAN) and reparameterization. Then we investigate how detector head size affects detection performance and find that a heavy neck with only one task projection layer would yield better results.In addition, AlignedOTA is proposed to solve the misalignment problem in label assignment. And a distillation schema is introduced to improve performance to a higher level. Based on these new techs, we build a suite of models at various scales to meet the needs of different scenarios. For general industry requirements, we propose DAMO-YOLO-T/S/M/L. They can achieve 43.6/47.7/50.2/51.9 mAPs on COCO with the latency of 2.78/3.83/5.62/7.95 ms on T4 GPUs respectively. Additionally, for edge devices with limited computing power, we have also proposed DAMO-YOLO-Ns/Nm/Nl lightweight models. They can achieve 32.3/38.2/40.5 mAPs on COCO with the latency of 4.08/5.05/6.69 ms on X86-CPU. Our proposed general and lightweight models have outperformed other YOLO series models in their respective application scenarios.

研究动机与目标

• 促进工业部署的实时对象检测改进。
• 开发面向延迟的神经架构搜索的基于 YOLO 的检测器。
• 设计高效的颈部与轻量化头部以优化延迟-准确性权衡。
• 改进标签分配，使分类与回归在动态设置下对齐。
• 融合蒸馏以提升较小模型的性能。

提出的方法

• 使用 MAE-NAS 进行延迟约束下的骨干网搜索，产生 ResNet-like/CSP-like 结构。
• 开发高效的 RepGFPN 颈部，在尺度间实现灵活的通道维度，并在 queen-fusion 中移除上采样。
• 引入 ZeroHead 仅保留一个任务投影层，使大颈部搭配小头部成为可能。
• 提出 AlignOTA，一种对齐的动态标签分配，平衡分类和回归成本。
• 应用蒸馏，结合两阶段训练和 Channel-wise Dynamic Temperature，以提升小模型的性能。

实验结果

研究问题

• RQ1 在延迟约束下，NAS 设计的骨干网如何提高实时检测器对 COCO mAP 的提升？
• RQ2 在实时约束下，哪种颈部/头部配置能实现最佳精度？
• RQ3 AlignOTA 是否在动态标签分配中改善了分类与回归的对齐，从而提升 YOLO 风格检测器的性能？
• RQ4 在不损害实时吞吐的前提下，蒸馏如何提升小型 DAMO-YOLO 模型的表现？

主要发现

• DAMO-YOLO-T/S/M/L 在 T4 GPU 上的延迟为 2.78/3.83/5.62/7.95 ms，对应 COCO 的 mAP 为 43.6/47.7/50.2/51.9。
• DAMO-YOLO-Ns/Nm/Nl 轻量模型在 X86-CPU 上的延迟为 4.08/5.05/6.69 ms，对应 COCO 的 mAP 为 32.3/38.2/40.5。
• MAE-NAS 骨干网（MAE-Res、MAE-CSP）在准确性/延迟权衡上超越 CSP-Darknet 基线，尤其在更大/更深的网络中。
• 大颈部/单投影头（ZeroHead）设计在减少计算量的同时仍能取得强劲性能。
• AlignOTA 相较于 ATSS/sOTA/TOOD 提升了标签分配效果，实现更高的 AP。
• 蒸馏（偏向 CWD）提升了小到中等 DAMO-YOLO 模型的性能。