Efficient detectors for edge devices are often optimized for metrics like parameters or speed counts, which remain weak correlation with the energy of detectors. However, among vision applications of convolutional neural networks (CNNs), some, such as always-on surveillance cameras, are critical for energy constraints. This paper aims to serve as a baseline by designing detectors to reach tradeoffs between energy and performance from two perspectives: 1) We extensively analyze various CNNs to identify low-energy architectures, including the selection of activation functions, convolutions operators, and feature fusion structures on necks. These underappreciated details in past works seriously affect the energy consumption of detectors; 2) To break through the dilemmatic energy-performance problem, we propose a balanced detector driven by energy using discovered low-energy components named \textit{FemtoDet}. In addition to the novel construction, we further improve FemtoDet by considering convolutions and training strategy optimizations. Specifically, we develop a new instance boundary enhancement (IBE) module for convolution optimization to overcome the contradiction between the limited capacity of CNNs and detection tasks in diverse spatial representations, and propose a recursive warm-restart (RecWR) for optimizing training strategy to escape the sub-optimization of light-weight detectors, considering the data shift produced in popular augmentations. As a result, FemtoDet with only 68.77k parameters achieves a competitive score of 46.3 AP50 on PASCAL VOC and power of 7.83W on RTX 3090. Extensive experiments on COCO and TJU-DHD datasets indicate that the proposed method achieves competitive results in diverse scenes.

翻译：为边缘设备设计的高效检测器通常针对参数量或速度等指标进行优化，但这些指标与检测器的能量消耗关联较弱。然而，在卷积神经网络（CNN）的视觉应用中，部分场景（如始终开启的监控摄像头）对能量约束至关重要。本文旨在通过从两个角度设计检测器来建立能量与性能之间的权衡基线：1）我们广泛分析多种CNN以识别低能耗架构，包括激活函数的选择、卷积算子以及颈部特征融合结构。这些以往工作中被忽视的细节严重影响了检测器的能耗；2）为突破能量-性能困境，我们利用发现的低能耗组件提出了一种能量驱动的平衡检测器，命名为FemtoDet。除新颖的架构设计外，我们进一步通过考虑卷积与训练策略优化来改进FemtoDet。具体而言，我们开发了一种新的实例边界增强（IBE）模块用于卷积优化，以克服CNN有限容量与检测任务在多样化空间表征之间的矛盾；同时提出递归热重启（RecWR）策略优化训练过程，以规避轻量级检测器因主流数据增强产生的分布偏移而陷入次优解。实验结果表明，FemtoDet仅凭68.77k参数量即可在PASCAL VOC上达到46.3 AP50的竞争性分数，在RTX 3090上功耗为7.83W。在COCO和TJU-DHD数据集上的大量实验表明，本方法在多样场景中均取得了具有竞争力的结果。