Research map / 3D vision / Face and shape
RayDF: Neural Ray-surface Distance Fields with Multi-view Consistency
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 multi-view visibility classification and consistency optimization to a ray-distance field (ablation) | RayDF · Blender, Group 1 | ADE (cm): 47.08 -> 7.97 | Helped |
| Use surface position (x,y,z) rather than ray distance d in k() (ablation) | RayDF visibility classifier · Blender | ADE (cm): 10.82 -> 7.97 | Helped |
| Add the surface-point encoder k() (ablation) | RayDF visibility classifier · Blender | ADE (cm): 10.23 -> 7.97 | Helped |
| Use symmetric average pooling rather than concatenating the two ray features (ablation) | RayDF visibility classifier · Blender | ADE (cm): 9.77 -> 7.97 | Helped |
| Increase sampled multi-view rays from 20 to 40 (ablation) | RayDF · Blender | ADE (cm): 7.97 -> 7.72 | Helped |
| Reduce sampled multi-view rays from 20 to 10 (ablation) | RayDF · Blender | ADE (cm): 7.97 -> 10.12 | Hurt |
| Train with only 1% of depth values (ablation) | RayDF · Blender | ADE (cm): 7.97 -> 8.54 | Hurt |
| Train both networks simultaneously rather than in two stages (ablation) | RayDF · Blender | ADE (cm): 7.97 -> 12.41 | Hurt |
| Replace the L1 consistency loss with L2 (ablation) | RayDF · Reception, DM-SR | ADE (cm): 6.96 -> 6.56 | Helped |
| Replace the L1 consistency loss with L2 (ablation) | RayDF · Reception, DM-SR | CD (×10⁻³) mean: 10.632 -> 12.141 | Hurt |
| Add noise with level σ²=1.0 to visibility scores (ablation) | RayDF visibility classifier · Blender | ADE (cm): 7.97 -> 17.03 | Hurt |
| Add noise with level σ²=0.5 to visibility scores (ablation) | RayDF visibility classifier · Blender | ADE (cm): 7.97 -> 15.54 | Hurt |
Closest papers by meaning
- RayNet: Learning Volumetric 3D Reconstruction With Ray Potentials (CVPR 2018 · similarity 0.90)
- FoundSurface: Feedforward Ray-Based 3D Scene Surface Reconstruction from Unposed Images (NeurIPS 2026 · similarity 0.89)
- DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling (CVPR 2024 · similarity 0.89)
- Ray Priors Through Reprojection: Improving Neural Radiance Fields for Novel View Extrapolation (CVPR 2022 · similarity 0.89)
- RaySt3R: Predicting Novel Depth Maps for Zero-Shot Object Completion (NeurIPS 2025 · similarity 0.88)
- ProbeSDF: Light Field Probes For Neural Surface Reconstruction (CVPR 2025 · 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.