Research map / 3D vision / Depth and stereo
Generalized Event Cameras
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Designs generalized event cameras on single-photon (SPAD) sensors by generalizing what an event transmits (integrated flux) and when it is triggered (Bayesian, patch-chunk, and coded-exposure change detectors), yielding intensity-preserving, bandwidth-efficient video.
Problem
Conventional event cameras transmit only binary brightness-change polarities, which lack the scene intensity needed for general vision tasks, so they cannot support plug-and-play downstream inference.
Approach
Parameterize event cameras along two axes, an integrator (what is transmitted, e.g., cumulative flux since last event) and a change detector (when transmitted), and realize multiple designs on digital SPAD photon detections: Bayesian online change-point detection, spatiotemporal patch-chunk events with a learned linear feature norm, and coded-exposure events with binomial confidence intervals.
Key results
- Spatiotemporal chunk events achieve 3025 FPS reconstruction at 431 bps/pixel, a 227x compression of raw SPAD capture.
- Generalized event cameras give a 4-8 dB PSNR improvement over EDI++ at equal readout and compress raw photon data by about 80x.
- On-chip UltraPhase implementations reduce bandwidth and power by over two orders of magnitude relative to raw photon readout.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace binary event polarities with n-bit integrator values (adaptive exposure) and a change detector | event camera · High-speed videography and rate-distortion | PSNR vs readout: 4-8 dB PSNR over EDI++ at same readout | Helped |
| Spatiotemporal chunk change detection (Sec 4.2) instead of other generalized designs (ablation) | single-photon generalized event camera · Rate-distortion analysis | PSNR | Helped |
| Bayesian change detector (Sec 4.1) instead of fixed-threshold EMA detector (ablation) | adaptive-EMA event camera · Qualitative change detection on hole saw bit | detection of low-contrast changes | Helped |
| Coded-exposure change detection instead of chunk or Bayesian designs (ablation) | single-photon generalized event camera · On-chip feasibility (UltraPhase) | compute and power cost | Helped |
| Bayesian change detector (Sec 4.1) instead of chunk-based or coded designs (ablation) | single-photon generalized event camera · On-chip feasibility (UltraPhase) | compute cost | Hurt |
| Generalized event cameras vs. raw SPAD photon readout on-chip | single-photon generalized event camera · UltraPhase bandwidth and power | bandwidth and power: over two orders of magnitude reduction | Helped |
Limitations
- Rigid dynamics scenes can be worse than burst photography with much higher readout.
- Modulated light sources such as phone screens trigger many unwanted events.
- Camera ego-motion triggers events in static regions, reducing bandwidth savings.
Experimental setup
Datasets: XVFI
Benchmarks: PSNR rate-distortion on YouTube Phantom Flex4k 1000 FPS videos tennis plug-and-play inference (HRNet, corner detection, RAFT, DETR, SAM)
Metrics: PSNR bits per second per pixel bandwidth power
Baselines: EDI++ E2VID+ Prophesee EVK4 events Compressive sensing with 8-bucket masks Burst denoising Long exposure Arc* E-RAFT
Builds on: BOCPD EDI E2VID+ Coded exposure photography
Closest papers by meaning
- Event Fields: Capturing Light Fields at High Speed, Resolution, and Dynamic Range (CVPR 2025 · similarity 0.87)
- Event-3DGS: Event-based 3D Reconstruction Using 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.85)
- GS2E: Gaussian Splatting is an Effective Data Generator for Event Stream Generation (NeurIPS 2025 · similarity 0.84)
- Seeing Motion at Nighttime with an Event Camera (CVPR 2024 · similarity 0.84)
- EventSplat: 3D Gaussian Splatting from Moving Event Cameras for Real-time Rendering (CVPR 2025 · similarity 0.84)
- Towards 3D Vision with Low-Cost Single-Photon Cameras (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.