Research map / 3D vision / Depth and stereo
Event Fields: Capturing Light Fields at High Speed, Resolution, and Dynamic Range
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.
Proposes Event Fields, capturing high-speed, high-resolution, HDR light fields with an event camera using kaleidoscope (spatial multiplexing) or galvanometer (temporal multiplexing) optics, validated by a Blender simulator and hardware prototypes.
Problem
Event cameras record brightness changes integrated over all angles, discarding angular (light field) information, so high-speed light field capture with post-capture refocusing and depth is not possible with standard event sensing.
Approach
Multiplex angular information into the event camera's spatial dimension (kaleidoscope mirror array, yielding angular temporal derivatives) or temporal dimension (galvanometer scanning a Lissajous/circular curve, yielding angular derivatives), and reconstruct light fields with pre-trained event-to-video (E2VID) and TimeLens sensor fusion.
Key results
- Galvanometer design at 250 Hz reconstructs light fields at 250 fps with 40 views at megapixel resolution, using E2VID.
- Kaleidoscope SloMoRF interpolates 120 fps RGB video to 720 fps (5x) with refocusing across three focal planes.
- Galvanometer scan frequency beyond 200-250 Hz causes event loss from readout bandwidth limits, blurring static objects in refocused images.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use kaleidoscope spatial multiplexing instead of galvanometer temporal multiplexing | Event Fields · Dynamic scenes | spatial resolution | Hurt |
| Use kaleidoscope spatial multiplexing instead of galvanometer temporal multiplexing | Event Fields · Static scenes | scene reconstruction | Hurt |
| Use galvanometer temporal multiplexing instead of kaleidoscope spatial multiplexing | Event Fields · Dynamic scenes | motion blur | Hurt |
| Increase galvanometer scan frequency | Galvanometer Event Fields · Refocusing on a rotating fan | text clarity | Helped |
| Increase galvanometer scan frequency beyond 250 Hz | Galvanometer Event Fields · Static foreground and background objects | sharpness: 200 Hz; 250 Hz | Hurt |
| Capture the light field with galvanometer temporal multiplexing | Event Fields · Dynamic scenes | refocusing capability | Helped |
| Combine kaleidoscope event capture with RGB sensor fusion | SloMoRF · Dynamic scenes | frame rate: from 120 to 720 fps | Helped |
| Use galvanometer-based event fields for depth-from-focus | Galvanometer Event Fields · Running person depth estimation | depth estimation rate: 100 fps | Helped |
Surprising findings
- Static objects become blurrier at higher galvo scan frequencies because the event camera hits its readout bandwidth limit and drops events.
- The kaleidoscope design cannot reconstruct static scene regions, since no events are generated without motion.
Limitations
- Kaleidoscope design yields lower spatial resolution and captures no static scene content.
- Galvanometer design suffers motion blur for fast-moving objects and event-camera bandwidth limits beyond 250 Hz.
- Relies on pre-trained E2VID and TimeLens models not trained for event field data; TimeLens++ code not publicly available.
Experimental setup
Datasets: Blender-simulated event fields captured kaleidoscope and galvanometer event data
Baselines: TimeLens++ Fujii et al. galvanometer light field Guo et al. microlens array event light field
Builds on: Manakov et al. kaleidoscope light field TimeLens E2VID depth from focus light field refocusing (Levoy and Hanrahan)
Closest papers by meaning
- Time-Efficient Light-Field Acquisition Using Coded Aperture and Events (CVPR 2024 · similarity 0.90)
- EventSplat: 3D Gaussian Splatting from Moving Event Cameras for Real-time Rendering (CVPR 2025 · similarity 0.88)
- Generalized Event Cameras (CVPR 2024 · similarity 0.87)
- Seeing Motion at Nighttime with an Event Camera (CVPR 2024 · similarity 0.87)
- Event-3DGS: Event-based 3D Reconstruction Using 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.87)
- Towards Robust Event-guided Low-Light Image Enhancement: A Large-Scale Real-World Event-Image Dataset and Novel Approach (CVPR 2024 · similarity 0.87)
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.