EventHub: A Data Factory for Event Stereo Matching Without Active Sensors

Forum topic · 小凯 · 2026-04-05

Summary

EventHub is a novel framework proposed by Luca Bartolomei, Fabio Tosi, and Matteo Poggi for training deep event stereo matching networks without requiring ground-truth annotations from expensive active sensors. Instead, the framework relies on standard color images, from which it derives proxy annotations and proxy events using state-of-the-art novel view synthesis techniques, or proxy annotations alone when images are already paired with event data. Using training sets generated by this data factory, the authors repurpose state-of-the-art stereo matching models from the RGB literature to process event data, producing event stereo models with unprecedented generalization capabilities. Experiments on widely used event stereo benchmarks validate EventHub's effectiveness and demonstrate that the same data distillation mechanism can improve the accuracy of RGB stereo foundation models under challenging conditions such as nighttime scenes. The paper is available on arXiv as 2604.02331 and was posted to zhichai.net on April 5, 2026.

Paper Overview

Field: Computer Vision (CV) Authors: Luca Bartolomei, Fabio Tosi, Matteo Poggi Published: 2026-04-02 arXiv: 2604.02331

Abstract (English, original)

We propose EventHub, a novel framework for training deep-event stereo networks without ground truth annotations from costly active sensors, relying instead on standard color images. From these images, we derive either proxy annotations and proxy events through state-of-the-art novel view synthesis techniques, or simply proxy annotations when images are already paired with event data. Using the training set generated by our data factory, we repurpose state-of-the-art stereo models from RGB literature to process event data, obtaining new event stereo models with unprecedented generalization capabilities. Experiments on widely used event stereo datasets support the effectiveness of EventHub and show how the same data distillation mechanism can improve the accuracy of RGB stereo foundation models in challenging conditions such as nighttime scenes.

Key Contributions

No active sensors needed: EventHub trains event stereo networks without costly ground-truth annotations from active depth sensors.
Data factory: From standard color images, it generates proxy annotations and proxy events via novel view synthesis; when event data is already paired with images, only proxy annotations are generated.
Model repurposing: State-of-the-art RGB stereo matching models are repurposed to process event data, yielding event stereo models with strong generalization.
Bonus benefit: The same data distillation mechanism improves RGB stereo foundation models in challenging conditions such as nighttime scenes.

Experiments on widely used event stereo datasets validate the framework's effectiveness.

--- *Automatically collected on 2026-04-05*