Research map / 3D vision / Depth and stereo
NeLU3D: Neural Inverse Structured Light without Modeling the Projector
Giancarlo Pereira, David Fouhey, Claudio Silva, Daniele Panozzo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ambient MLP with Iambient<=Iwhite constraint (sigmoid times Iwhite) (ablation) | NeLU3D · pawn (ablation) | MAE: unconstrained ambient: FAILED (exploding gradients); full model MAE: 0.65 mm | Improved |
| add ambient MLP a on top of lookup-only loss (ablation) | NeLU3D · pawn (ablation) | MAE: only LookUp fails to model images; full model MAE: 0.65 mm | Improved |
| add lookup loss to pattern-only loss (ablation) | NeLU3D · pawn (ablation) | MAE: only pattern still fails to overcome global illumination; full MAE: 0.65 mm | Improved |
| replace TurboSL with NeLU3D on 3 Random patterns | TurboSL · monkey, 3 random patterns | mean absolute depth error (mm): 40.04 -> 0.96 | Improved |
| replace TurboSL with NeLU3D on 3 Random patterns | TurboSL · first aid, 3 random patterns | mean absolute depth error (mm): 122.00 -> 0.53 | Improved |
| replace TurboSL with NeLU3D on 3 Hilbert patterns | TurboSL · wipes, 3 Hilbert patterns | mean absolute depth error (mm): 29.91 -> 3.87 | Improved |
| replace TurboSL with NeLU3D on 3 a la carte patterns | TurboSL · teapot, 3 a la carte patterns | mean absolute depth error (mm): 2.90 -> 9.17 | Hurt |
| replace TurboSL with NeLU3D on 4 MPS patterns | TurboSL · pawn, 4 MPS patterns | mean absolute depth error (mm): 2.76 -> 1.02 | Improved |
| replace ZNCC with NeLU3D on 11 Gray codes | ZNCC · keyboard, 11 Gray codes | mean absolute depth error (mm): 0.39 -> 0.33 | No clear effect |
| replace TurboSL with NeLU3D (reconstruction time) | TurboSL · pawn, k=3 patterns | time (s): 1020 -> 420 | Improved |
| replace TurboSL with NeLU3D (reconstruction time) | TurboSL · pawn, k=11 patterns | time (s): 2364 -> 431 | Improved |
| replace TurboSL with NeLU3D (memory) | TurboSL · pawn, k=11 patterns | VRAM (GB): 5.2 -> 4.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TurboSL — TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024) · better on 9 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024 · similarity 0.88)
- FreeCam3D: Snapshot Structured Light 3D with Freely-Moving Cameras (ECCV 2020 · similarity 0.87)
- Real-time Acquisition and Reconstruction of Dynamic Volumes with Neural Structured Illumination (CVPR 2024 · similarity 0.85)
- NeRFPrior: Learning Neural Radiance Field as a Prior for Indoor Scene Reconstruction (CVPR 2025 · similarity 0.85)
- NTO3D: Neural Target Object 3D Reconstruction with Segment Anything (CVPR 2024 · similarity 0.85)
- From Transparent to Opaque: Rethinking Neural Implicit Surfaces with $\alpha$-NeuS (NeurIPS 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.
