Generative Anonymization in Event Streams

Adam T. Müller; Mihai Kocsis; Nicolaj C. Stache

cs.CVApr 14, 2026

Generative Anonymization in Event Streams

Authors: Adam T. Müller, Mihai Kocsis, Nicolaj C. Stache

Organizations: Heilbronn University of Applied Sciences, Germany

Abstract

Neuromorphic vision sensors offer low latency and high dynamic range, but their deployment in public spaces raises severe data protection concerns. Recent Event-to-Video (E2V) models can reconstruct high-fidelity intensity images from sparse event streams, inadvertently exposing human identities. Current obfuscation methods, such as masking or scrambling, corrupt the spatio-temporal structure, severely degrading data utility for downstream perception tasks. In this paper, to the best of our knowledge, we present the first generative anonymization framework for event streams to resolve this utility-privacy trade-off. By bridging the modality gap between asynchronous events and standard spatial generative models, our pipeline projects events into an intermediate intensity representation, leverages pretrained models to synthesize realistic, non-existent identities, and re-encodes the features back into the neuromorphic domain. Experiments demonstrate that our method reliably prevents identity recovery from E2V reconstructions while preserving the structural data integrity required for downstream vision tasks. Finally, to facilitate rigorous evaluation, we introduce a novel, synchronized real-world event and RGB dataset captured via precise robotic trajectories, providing a robust benchmark for future research in privacy-preserving neuromorphic vision.

Figures & tables

Figure 1: Architectural overview of the generative anonymization pipeline. The framework translates raw asynchronous event data into continuous grayscale frames to detect and swap faces using established generative models. The anonymized identity is subsequently projected back into the event space via a V2E conversion, preserving the underlying spatiotemporal structure.

Figure 2: Qualitative examples of source and synthetic identities. Comparison of three subjects (rows). Columns from left-to-right: Source event streams, intermediate E2V representations, the anonymized generative output (Anon), V2E projection into a new event stream, and the final downstream E2V validation.

Metric	Identity Similarity $\downarrow$		Temporal Stability $\uparrow$		Pose Error $\downarrow$ / °		Mimicry Error $\downarrow$
	Anonymization	Reference	Anonymization	Reference	Anonymization	Reference	Anonymization	Reference
Mean $\mu$	0.118	0.713	0.760	0.770	3.304	2.613	0.181	0.239
Std. Deviation $\sigma$	0.0172	0.0159	0.0135	0.0325	0.453	0.566	0.0298	0.1474

Table 1: Anonymization metrics and feature preservation in image space. Anonymization compares the source stream to our generative output. Reference compares two independent source streams of the same subject, establishing the natural variance limit for identity and pose.

Metric	YOLO Conf. Anonymization $\uparrow$	YOLO Conf. Reference $\uparrow$	YOLO IoU $\uparrow$	YOLO Det.-Rate Error $\downarrow$	Event IoU $\uparrow$
Mean $\mu$	0.894	0.937	0.960	0.000	0.702
Std. Deviation $\sigma$	0.011	0.007	0.010	0.000	0.137

Table 2: Downstream task performance for face detection. Intensity-domain ( YOLO ) and event-domain ( Event ) metrics are computed using YOLOv8 [ 39 ] and an event-based detector [ 9 ] , respectively. IoU and detection rate error compare unmodified source data ( Reference ) against the anonymized output ( Anonymization ).

		Mean $\mu$	Std. Deviation $\sigma$
STCD $\uparrow$	Anonymization	0.3143	0.0415
	Reference	0.0099	0.0029
EMD $\uparrow$	Anonymization	0.1276	0.0132
	Reference	0.0085	0.0039

Table 3: Structural anonymization metrics in the event space. Anonymization compares source and synthetic streams, while Reference evaluates two independent captures of the same subject to provide a baseline geometric distance.

Figure 3: V2E discretization and density artifacts. Viewed tilted from the top-down position, where more recent events are closer to the frontal cross-section. The reverse projection step relies on standard V2E conversion, which leads to discretization in the event-space. Notably only in the parts of the event stream where information has been replaced (facial region).

Event-Based Vision

Event-based vision uses asynchronous camera measurements to perceive motion and scene changes with high temporal resolution and dynamic range.

See all 232 papers

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision
Jun 18, 2026Roberto Pellerito, Daniel Gehrig, Shintaro Shiba +1Event-Based Vision Representation Learning
Robotics and Perception Group, University of Zurich · University of Pennsylvania · The University of Tokyo +1
Towards Video Anomaly Detection from Event Streams: A Baseline and Benchmark Datasets
Mar 26, 2026Peng Wu, Yuting Yan, Guansong Pang +4Event-Based Vision Video Anomaly Detection
School of Computer Science, Northwestern Polytechnical University, China · School of Computing and Information Systems, Singapore Management University, Singapore · School of Artificial Intelligence, Xidian University, China
Event-to-Video Reconstruction using Spatio-Temporal and Frequency-Enhanced Deep Neural Networks
May 25, 2026Ramna Maqsood, Paulo Nunes, Luís Ducla Soares +1Event-Based Vision Vision Transformer
Instituto de Telecomunicações, Instituto Universitário de Lisboa (ISCTE-IUL) Lisbon, Portugal · Instituto de Telecomunicac¸˜oes, Instituto Universit´ario de Lisboa (ISCTE-IUL) Lisbon, Portugal

Jun 18, 2026J/K move · Enter open · S save

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision

Roberto Pellerito, Daniel Gehrig, Shintaro Shiba, Davide Scaramuzza

Robotics and Perception Group, University of Zurich · University of Pennsylvania · The University of Tokyo+1

Event-Based Vision Representation Learning Autoencoders Event-Based Object Detection

Open paper AskPDF arXiv

Event cameras capture dynamic scenes with exceptional temporal fidelity by representing them as a continuous stream of microsecond resolution events. Each individual event, however, only carries minimal semantic value, merely signaling a localized brightness change. To derive meaningful signals, downstream algorithms need to quickly integrate cues from a potentially massive torrent of low-information events. Current architectures, however, are easily overwhelmed, struggling to balance capturing fine-grained temporal dynamics and maintaining a manageable data throughput. This paper proposes a framework to re-tokenize event streams into a small set of highly informative neural events, each representing a local spatio-temporal context window with a discrete learnable code. Every time this code flips, a neural event is triggered, yielding a highly compressed data stream. We demonstrate that, across object detection and classification, networks trained on neural events are on par or surpass the performance of state-of-the-art approaches while reducing the event rate by a factor of 2.0.

Abstract

Event-Based Vision

Explore similar work

Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision

Towards Video Anomaly Detection from Event Streams: A Baseline and Benchmark Datasets

Event-to-Video Reconstruction using Spatio-Temporal and Frequency-Enhanced Deep Neural Networks