CED: Color Event Camera Dataset

Event: CVPR 2019 Workshop · Duration: 3 min · ▶ Watch on YouTube

Abstract

This presentation introduces CED, the first color event camera dataset, which provides 50 minutes of synchronized DAVIS frames and color events. The dataset captures a wide variety of scenes, including indoor, outdoor, and driving scenarios, under diverse lighting conditions ranging from low light (0.8 lux) to direct sunlight (10,000 lux), and features high dynamic range and fast 6-DOF motions. The speaker also demonstrates how multiple grayscale image reconstruction methods can be adapted to visualize the color event stream, highlighting the advantages of color event cameras in challenging environments. The dataset is made publicly available to encourage further research and development in event-based vision.

Speakers

  • Henri Rebecq — University of Zurich, ETH Zurich
  • Cedric Scheerlinck — Australian National University
  • Timo Stoffregen — University of Zurich, ETH Zurich
  • Nick Barnes — Australian National University
  • Robert Mahony — Australian National University
  • Davide Scaramuzza — University of Zurich, ETH Zurich

Talks (1)

  • 00:00:00 — Henri Rebecq: CED: Color Event Camera Dataset
    • Introduction to the Color Event Camera Dataset (CED), its features, sample sequences, and event stream visualization methods.

Key Takeaways

  • CED is the first color event camera dataset, offering 50 minutes of synchronized DAVIS frames and color events.
  • The dataset covers a wide range of challenging conditions, including diverse scenes, extreme lighting, high dynamic range, and fast motions.
  • Color event cameras provide rich color information and excel in conditions where traditional cameras struggle, such as high-speed or HDR scenarios.
  • Multiple grayscale image reconstruction methods have been adapted to visualize color event streams, demonstrating their potential for various applications.
  • The dataset is publicly available to foster research and development in event-based vision.

Methods / Models / Datasets Mentioned

  • DAVIS346 Red Color
  • Bayer pattern
  • Manifold Regularization
  • High-pass Filter
  • Neural Network (E2VID)
  • Taverni et al., Front and back Illuminated Dynamic and Active Pixel Vision Sensors comparison, TCS'18
  • Scheerlinck et al., CED: Color Event Camera Dataset, CVPR-W'19
  • Munda et al., UCV'18
  • Scheerlinck et al., ACCV'18
  • Rebecq et al., CVPR'19

Topics

Color event cameras · Event-based vision · Dataset · DAVIS camera · High dynamic range · Low light · High speed motion · Image reconstruction · Color information · Robotics

Notes

Open for commentary — connections to other work, critiques, follow-up reading.