OCDet: Object Center Detection via Bounding Box-Aware Heatmap Prediction on Edge Devices with NPUs

TL;DR

This work introduces OCDet, a lightweight Object Center Detection framework optimized for edge devices with NPUs, and introduces Generalized Centerness (GC) to generate ground truth heatmaps from bounding box annotations, providing finer spatial details without additional manual labeling.

Abstract

Real-time object localization on edge devices is fundamental for numerous applications, ranging from surveillance to industrial automation. Traditional frameworks, such as object detection, segmentation, and keypoint detection, struggle in resource-constrained environments, often resulting in substantial target omissions. To address these challenges, we introduce OCDet, a lightweight Object Center Detection framework optimized for edge devices with NPUs. OCDet predicts heatmaps representing object center probabilities and extracts center points through peak identification. Unlike prior methods using fixed Gaussian distribution, we introduce Generalized Centerness (GC) to generate ground truth heatmaps from bounding box annotations, providing finer spatial details without additional manual labeling. Built on NPU-friendly Semantic FPN with MobileNetV4 backbones, OCDet models are trained by our Balanced Continuous Focal Loss (BCFL), which alleviates data imbalance and focuses training on hard negative examples for probability regression tasks. Leveraging the novel Center Alignment Score (CAS) with Hungarian matching, we demonstrate that OCDet consistently outperforms YOLO11 in object center detection, achieving up to 23% higher CAS while requiring 42% fewer parameters, 34% less computation, and 64% lower NPU latency. When compared to keypoint detection frameworks, OCDet achieves substantial CAS improvements up to 186% using identical models. By integrating GC, BCFL, and CAS, OCDet establishes a new paradigm for efficient and robust object center detection on edge devices with NPUs. The code is released at https://github.com/chen-xin-94/ocdet.

Figures (10)

• Figure 1: Performance comparison between our proposed OCDet framework and state-of-the-art real-time object detectors YOLOv8 and YOLO11, as well as a standard keypoint detection framework (KP) using identical model configurations. Results demonstrate OCDet’s Pareto dominance in latency-accuracy trade-offs.
• Figure 2: Illustration of our proposed OCDet, an Object Center Detection framework, comprising: ground truth generation leveraging Generalized Centerness (GC), model training through Balanced Continuous Focal Loss (BCFL), inference with NPU-oriented OCDet models, and evaluation using Hungarian matching and Center Alignment Score (CAS).
• Figure 3: Illustration of (a) $\alpha_\textrm{c}(y)$ given a certain $\alpha$ and (b) BCFL vs QFL under $\alpha=0.75$ across various target values $y \in [0,1]$.
• Figure 4: Illustration of Generalized Centerness heatmaps of a square bounding box under different $\eta$ and $\phi$ configurations.
• Figure 5: Evaluation of the proposed OCDet against state-of-the-art real-time object detectors YOLOv8, YOLO11, and the same models trained under a standard keypoint-based framework (KP). Corresponds to \ref{['fig:cas_latency']} in the main paper.