ComPrivDet: Efficient Privacy Object Detection in Compressed Domains Through Inference Reuse

Yunhao Yao; Zhiqiang Wang; Ruiqi Li; Haoran Cheng; Puhan Luo; Xiangyang Li

cs.CVApr 4, 2026

ComPrivDet: Efficient Privacy Object Detection in Compressed Domains Through Inference Reuse

Authors: Yunhao Yao, Zhiqiang Wang, Ruiqi Li, Haoran Cheng, Puhan Luo, Xiangyang Li

Organizations: University of Science and Technology of China (USTC)

Abstract

As the Internet of Things (IoT) becomes deeply embedded in daily life, users are increasingly concerned about privacy leakage, especially from video data. Since frame-by-frame protection in large-scale video analytics (e.g., smart communities) introduces significant latency, a more efficient solution is to selectively protect frames containing privacy objects (e.g., faces). Existing object detectors require fully decoded videos or per-frame processing in compressed videos, leading to decoding overhead or reduced accuracy. Therefore, we propose ComPrivDet, an efficient method for detecting privacy objects in compressed video by reusing I-frame inference results. By identifying the presence of new objects through compressed-domain cues, ComPrivDet either skips P- and B-frame detections or efficiently refines them with a lightweight detector. ComPrivDet maintains 99.75% accuracy in private face detection and 96.83% in private license plate detection while skipping over 80% of inferences. It averages 9.84% higher accuracy with 75.95% lower latency than existing compressed-domain detection methods.

Figures & tables

Fig. 1: The System Overview of ComPrivDet.

Fig. 2: Examples of Accumulated Motion Vectors and Accumulated Residuals.

$\tau_{ab}$	4e-3			8e-3			1.2e-2			1.6e-2			2e-2
$\tau_{conf}$	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat
0.1	0.9569		63.87	0.9661		53.82	0.9811		40.16	0.9894		27.83	0.9975		23.55
0.2	0.9564		61.44	0.9656		54.02	0.9750		39.60	0.9811		26.81	0.9889		23.45
0.3	0.9564	26.22%	59.24	0.9656	41.58%	52.14	0.9750	65.06%	36.48	0.9811	77.83%	24.43	0.9889	84.75%	21.31
0.4	0.9564		57.98	0.9656		50.56	0.9750		35.63	0.9811		24.35	0.9889		20.38
0.5	0.9561		56.63	0.9644		46.62	0.9689		32.76	0.9731		24.59	0.9806		19.55

TABLE I: Evaluation of Accuracy, Reuse Ratio, Latency (ms) under Different

\tau_{conf}

\tau_{ab}

Settings for Private Face Detection.

$\tau_{ab}$	4e-3			8e-3			1.2e-2			1.6e-2			2e-2
$\tau_{conf}$	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat	Acc	RR	Lat
0.1	0.9172		61.18	0.9219		55.92	0.9256		36.16	0.9442		27.74	0.9600		23.09
0.2	0.9256		61.16	0.9303		51.77	0.9339		36.53	0.9525		26.99	0.9683		21.62
0.3	0.9250	26.22%	59.67	0.9264	41.58%	48.29	0.9300	65.06%	35.58	0.9442	77.83%	25.91	0.9600	84.75%	22.28
0.4	0.9167		58.07	0.9181		48.42	0.9217		35.71	0.9358		25.47	0.9517		19.22
0.5	0.9167		56.03	0.9181		47.34	0.9217		33.57	0.9358		25.49	0.9517		19.93

TABLE II: Evaluation of Accuracy, Reuse Ratio, Latency (ms) under Different

\tau_{conf}

\tau_{ab}

Settings for Private License Plate Detection.

Panel of Figure 5 — Fig. 3: Evaluation of Abnormality Determination Latency Ratio.

Panel of Figure 6 — Fig. 4: Comparison with Existing Pixel-Domain Detectors

Panel of Figure 7 — Fig. 5: Comparison with Existing Compressed-Domain Frame-Level Detectors

Panel of Figure 8 — Fig. 6: Comparison with Existing Compressed-Domain GOP-Level Detectors

Object Detection

Object detection locates and classifies objects in images or video, balancing localization accuracy, robustness, and inference speed.

See all 624 papers

Only Whats Necessary: Pareto Optimal Data Minimization for Privacy Preserving Video Anomaly Detection
Mar 27, 2026Nazia Aslam, Abhisek Ray, Thomas B. Moeslund +1Privacy-Preserving ML Privacy-Utility Tradeoff
Aalborg University, Denmark · Aarhus University, Denmark · Milestone System, Denmark
MVTrack: Ultrafast Appearance-Free Moving Object Tracking from Compressed Bitstreams
Aug 11, 2026Iñaki Erregue, Kamal Nasrollahi, Sergio EscaleraVisual Object Tracking Video Object Detection
Universitat de Barcelona · Computer Vision Center · Milestone Systems +1
Privacy-Aware Video Anomaly Detection through Orthogonal Subspace Projection
May 9, 2026Lei Wang, Wenxiang Diao, Andrew Busch +2Privacy-Preserving ML Video Anomaly Detection
School of Engineering and Built Environment, Griffith University · Data61/CSIRO · School of Computer Science and Engineering, University of New South Wales +1

Mar 27, 2026J/K move · Enter open · S save

Only Whats Necessary: Pareto Optimal Data Minimization for Privacy Preserving Video Anomaly Detection

Nazia Aslam, Abhisek Ray, Thomas B. Moeslund, Kamal Nasrollahi

Aalborg University, Denmark · Aarhus University, Denmark · Milestone System, Denmark

Privacy-Preserving ML Privacy-Utility Tradeoff Video Anomaly Detection

Open paper AskPDF arXiv

Video anomaly detection (VAD) systems are increasingly deployed in safety critical environments and require a large amount of data for accurate detection. However, such data may contain personally identifiable information (PII), including facial cues and sensitive demographic attributes, creating compliance challenges under the EU General Data Protection Regulation (GDPR). In particular, GDPR requires that personal data be limited to what is strictly necessary for a specified processing purpose. To address this, we introduce Only What's Necessary, a privacy-by-design framework for VAD that explicitly controls the amount and type of visual information exposed to the detection pipeline. The framework combines breadth based and depth based data minimization mechanisms to suppress PII while preserving cues relevant to anomaly detection. We evaluate a range of minimization configurations by feeding the minimized videos to both a VAD model and a privacy inference model. We employ two ranking based methods, along with Pareto analysis, to characterize the resulting trade off between privacy and utility. From the non-dominated frontier, we identify sweet spot operating points that minimize personal data exposure with limited degradation in detection performance. Extensive experiments on publicly available datasets demonstrate the effectiveness of the proposed framework.

Abstract

Object Detection

Explore similar work

Only Whats Necessary: Pareto Optimal Data Minimization for Privacy Preserving Video Anomaly Detection

MVTrack: Ultrafast Appearance-Free Moving Object Tracking from Compressed Bitstreams

Privacy-Aware Video Anomaly Detection through Orthogonal Subspace Projection