Research map / 3D vision / Face and shape
Approximate Differentiable Rendering with Algebraic Surfaces
Leonid Keselman, Martial Hebert
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 |
|---|---|---|---|
| use Fuzzy Metaballs renderer instead of SoftRas mesh renderer | pose estimation · Pose estimation (noisy depth) | pose error: 17.0 -> 4.2 | Improved |
| use Fuzzy Metaballs instead of high-fidelity point ICP | pose estimation · Pose estimation (noisy depth) | pose error: 6.8 -> 4.2 | Improved |
| use Fuzzy Metaballs instead of equal-fidelity point ICP | pose estimation · Pose estimation (noise-free) | pose error: 7.6 -> 4.0 | Improved |
| use Fuzzy Metaballs instead of Pulsar | pose estimation · Pose estimation (noise-free) | pose error: 20.2 -> 4.0 | Improved |
| use Fuzzy Metaballs instead of NeRF | shape from silhouette · Shape from silhouette (noisy) | silhouette cross-entropy: 0.062 -> 0.055 | Improved |
| use NeRF instead of Fuzzy Metaballs | shape from silhouette · Shape from silhouette (noise-free) | silhouette cross-entropy: 0.040 -> 0.032 | Hurt |
| use Fuzzy Metaballs instead of PyTorch3D mesh | shape from silhouette · Shape from silhouette (runtime) | time (s): 3008 -> 68 | Improved |
| use Fuzzy Metaballs instead of voxel carving under noise | shape from silhouette · Shape from silhouette (noisy) | silhouette cross-entropy: 1.119 ± 0.367 -> 0.055 | Improved |
| use Fuzzy Metaballs instead of SoftRas on CPU | rendering · Rendering runtime (CPU, backward) | ms: 384.7 -> 13.2 | Improved |
| lower alpha threshold from 50% to 10% (ablation) | Fuzzy Metaballs · CO3D / phone video reconstruction | qualitative recovery | Improved |
| tune PyTorch3D hyper-parameters with CMA-ES (ablation) | PyTorch3D SoftRas baseline · pose estimation single frame | pose error: 16 degrees of rotation error and 9 percent of translation error | No clear effect |
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.
- Pulsar — Pulsar: Efficient Sphere-Based Neural Rendering (CVPR 2021) · better on 5 of 5 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
- Soft Rasterizer: A Differentiable Renderer for Image-Based 3D Reasoning (ICCV 2019 · similarity 0.90)
- A General Differentiable Mesh Renderer for Image-Based 3D Reasoning (TPAMI 2020 · similarity 0.89)
- Differentiable rendering with perturbed optimizers (NeurIPS 2021 · similarity 0.89)
- GenDR: A Generalized Differentiable Renderer (CVPR 2022 · similarity 0.88)
- Volumetric Surfaces: Representing Fuzzy Geometries with Layered Meshes (CVPR 2025 · similarity 0.86)
- LinPrim: Linear Primitives for Differentiable Volumetric Rendering (NeurIPS 2025 · 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.
