Research map / 3D vision / Depth and stereo
PS-EIP: Robust Photometric Stereo Based on Event Interval Profile
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 PS-EIP, a non-deep-learning event-based photometric stereo method that fits per-pixel surface normals to a time-series event interval profile and removes shadow/specular outliers with shape-based masks.
Problem
Event-based photometric stereo (EventPS) treats each event interval independently, making normal recovery sensitive to noise, shadows, and non-Lambertian reflections.
Approach
Model the inverse event intervals as a time-series Event Interval Profile (time derivative of log intensity), fit the theoretical profile per pixel to recover normals, and exclude outlier temporal regions (collapsed events, specularity, cast shadow) using shape-based masks.
Key results
- PS-EIP achieves an average MAE of 8.12 degrees over all 3D-printed objects versus 13.66 degrees for EventPS.
- On sphere-based shapes, average MAE is 6.72 degrees for the proposed method versus 11.87 degrees for EventPS.
- On complex shapes, average MAE is 9.66 degrees for the proposed method versus 15.64 degrees for EventPS.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use an event-interval time-series profile instead of treating intervals independently | PS-EIP · Sphere-based objects | MAE: 11.87° -> 6.72° | Helped |
| Use an event-interval time-series profile instead of treating intervals independently | PS-EIP · Complex shapes | MAE: 15.64° -> 9.66° | Helped |
| Use an event-interval time-series profile instead of treating intervals independently | PS-EIP · All 3D-printed objects | MAE: 13.66° -> 8.12° | Helped |
| Add shape-based masks to exclude distorted event-profile regions (ablation) | PS-EIP · Glossy, Pole, and MulRefs | MAE | Helped |
| Increase the number of measurements averaged (ablation) | PS-EIP · Diffuse, Glossy, and Pole | MAE | Helped |
| Increase averaging beyond 30 measurements (ablation) | PS-EIP · Diffuse, Glossy, and Pole | MAE | No clear effect |
| Increase ring-light rotational frequency (ablation) | PS-EIP · Diffuse | MAE | Mixed |
| Apply azimuth-offset correction (ablation) | PS-EIP · Diffuse | MAE | Helped |
Surprising findings
- Robust results are obtained without deep learning, outperforming the deep-learning variant EventPS-FCN.
Limitations
- Assumes distant light source and orthographic camera.
- Inter-reflections and subsurface scattering are not modeled, causing errors in concave regions.
- Constant event thresholds assumed; threshold variation affects profiles and requires averaging over 30 measurements.
Experimental setup
Datasets: real event data from 21 3D-printed objects
Benchmarks: 3D-printed known-shape objects (sphere-based and complex shapes)
Metrics: mean angular error (MAE)
Baselines: EventPS-FCN EventPS
Builds on: EventPS Woodham photometric stereo
Closest papers by meaning
- EventPS: Real-Time Photometric Stereo Using an Event Camera (CVPR 2024 · similarity 0.91)
- EventPSR: Surface Normal and Reflectance Estimation from Photometric Stereo Using an Event Camera (CVPR 2025 · similarity 0.87)
- Differentiable Display Photometric Stereo (CVPR 2024 · similarity 0.84)
- MVCPS-NeuS: Multi-view Constrained Photometric Stereo for Neural Surface Reconstruction (CVPR 2024 · similarity 0.83)
- SfPUEL: Shape from Polarization under Unknown Environment Light (NeurIPS 2024 · similarity 0.82)
- EAG3R: Event-Augmented 3D Geometry Estimation for Dynamic and Extreme-Lighting Scenes (NeurIPS 2025 · similarity 0.82)
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.