Research map / 3D vision / Depth and stereo
EventPSR: Surface Normal and Reflectance Estimation from Photometric Stereo Using an Event Camera
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 EventPSR, which uses an event camera under a moving ring light pattern and a two-stage (grid-matching then gradient-tuning) algorithm to jointly estimate surface normal, metallic and roughness with about 30% of the data rate of frame-based HDR methods.
Problem
Simultaneously recovering surface normal and reflectance parameters (roughness, metallic) normally requires many HDR frames from frame-based cameras, which is slow and data-heavy, especially for shiny materials.
Approach
Uses an event camera with a designed moving ring-light scanning pattern (chosen for specular coverage, diffuse sensitivity and event efficiency) and a per-pixel matching of event streams to a rendered database of intensity-time curves, refined by gradient-based tuning with interpixel smoothness.
Key results
- On synthetic data, EventPSR uses about 30% of the data rate of frame-based methods while achieving comparable normal error and superior roughness and metallic error.
- On real objects, EventPSR reaches an average roughness MSE of 0.0067 on semi-sphere roughness tests.
- Parameter recovery takes about 10 seconds per object on a single RTX 3060 GPU per the abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use EventPSR instead of the frame-based comparison methods | synthetic data, 16-object average · Synthetic data | data consumption ratio: 0.303 | Helped |
| use EventPSR instead of NeILF | normal estimation · Synthetic data, 16-object average | mean angular error: 9.516 -> 7.471 | Helped |
| use EventPSR instead of SDM-UniPS | normal estimation · Synthetic data, 16-object average | mean angular error: 9.516 -> 7.471 | Helped |
| use EventPSR instead of SDM-UniPS | roughness estimation · Synthetic data, 16-object average | roughness MSE: 0.015 -> 0.003 | Helped |
| use EventPSR instead of SDM-UniPS | metallic estimation · Synthetic data, 16-object average | metallic MSE: 0.048 -> 0.014 | Helped |
| estimate roughness with EventPSR | real semi-spheres with distinct roughness levels · Real data, roughness evaluation | roughness MSE: 0.0067 | Helped |
Limitations
- Capture speed is limited by monitor refresh rate.
- Real-object accuracy is affected by cast shadows and inter-reflection.
- Diffuse albedo cannot be recovered.
Experimental setup
Datasets: sculpture dataset DiLiGenT10
Benchmarks: synthetic rendered objects (16 sculptures) real objects captured on custom platform
Metrics: mean angular error mean squared error relative data consumption
Baselines: NeILF SDM-UniPS
Builds on: EventPS Disney Principled BRDF photometric stereo
Closest papers by meaning
- EventPS: Real-Time Photometric Stereo Using an Event Camera (CVPR 2024 · similarity 0.92)
- PS-EIP: Robust Photometric Stereo Based on Event Interval Profile (CVPR 2025 · similarity 0.87)
- SfPUEL: Shape from Polarization under Unknown Environment Light (NeurIPS 2024 · similarity 0.83)
- Event-3DGS: Event-based 3D Reconstruction Using 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.83)
- EvDiG: Event-guided Direct and Global Components Separation (CVPR 2024 · similarity 0.83)
- Color Event Enhanced Single-Exposure HDR Imaging (AAAI 2024 · similarity 0.83)
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.