Research map / 3D vision / Face and shape
Differentiable Surface Rendering via Non-Differentiable Sampling
Forrester Cole, Kyle Genova, Avneesh Sud, Daniel Vlasic, Zhoutong Zhang
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 surface geometry optimization (ablation) | SNeRF · NeRF Lego | PSNR: 27.44 -> 29.61 (improves 2.2 PSNR) | Improved |
| add surface geometry optimization (ablation) | SNeRF · NeRF Fern | PSNR: 23.41 -> 23.52 (improves only 0.1 PSNR) | No clear effect |
| replace volume rendering with surface light field rendering | NeRF · NeRF Lego | PSNR: 29.91 -> 29.61 | Hurt |
| replace volume rendering with surface light field rendering | NeRF · NeRF Fern | PSNR: 24.38 -> 23.52 (loses 0.9 PSNR) | Hurt |
| render a surface light field with one NeRF evaluation per pixel | NeRF · NeRF Lego | per-pixel cost (MFLOPS): 226 MFLOPS -> 1.7 MFLOPS (128 evaluations to one) | Improved |
| add sigma=0.5 splat blur (mipmapping on) | RtS · Texture optimization of textured sphere (mipmapping on) | PSNR: 33.7 -> 33.2 | Hurt |
| add sigma=0.5 splat blur (mipmapping off) | RtS · Texture optimization of textured sphere (mipmapping off) | PSNR: 25.4 -> 28.3 | Improved |
| use RtS-pose variant instead of RtS | RtS · Silhouette pose estimation (2M triangles) | ms per iteration: 26 -> 17 | Improved |
| use RtS instead of PyTorch3D | PyTorch3D · Silhouette pose estimation (2M triangles) | ms per iteration: 588 -> 26 (up to 20x speedup) | Improved |
| add multi-layer splatting (ablation) | RtS single-layer · Hand mesh optimization | convergence/stability | Improved |
| add multi-layer splatting (ablation) | RtS single-layer · Pose estimation of toy airplane | convergence | No clear effect |
| add splatting to rasterization (ablation) | RtS without splatting · Pose estimation of toy airplane | convergence | Improved |
Closest papers by meaning
- A Probability-guided Sampler for Neural Implicit Surface Rendering (ECCV 2024 · similarity 0.90)
- Differentiable Volumetric Rendering: Learning Implicit 3D Representations Without 3D Supervision (CVPR 2020 · similarity 0.89)
- A General Differentiable Mesh Renderer for Image-Based 3D Reasoning (TPAMI 2020 · similarity 0.89)
- UNISURF: Unifying Neural Implicit Surfaces and Radiance Fields for Multi-View Reconstruction (ICCV 2021 · similarity 0.89)
- Differentiable Neural Surface Refinement for Modeling Transparent Objects (CVPR 2024 · 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.
