Research map / 3D vision / Face and shape
DDSL: Deep Differentiable Simplex Layer for Learning Geometric Signals
Chiyu, Dana Lansigan, Philip Marcus, Matthias Niessner
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 DDSL rasterization loss to Chamfer-based mesh estimation (ablation) | single-image mesh estimation (ResNet18 + spherical conv) · ShapeNet airplane single-image mesh estimation | Chamfer distance (x10^-2): 9.16 -> 1.99 | Improved |
| add DDSL rasterization loss to Chamfer-based mesh estimation (ablation) | single-image mesh estimation · ShapeNet airplane single-image mesh estimation | accuracy (x10^-2): 8.47 -> 2.15 | Improved |
| add DDSL rasterization loss to Chamfer-based mesh estimation (ablation) | single-image mesh estimation · ShapeNet airplane single-image mesh estimation | completeness (x10^-2): 9.84 -> 1.83 | Improved |
| use analytic DDSL derivative instead of finite difference | DDSL backward pass · random simplex meshes, 5-50 points, resolution 4-32 | backward runtime: 10~100x faster | Improved |
| replace Polygon-RNN++ with DDSL-based PolygonNet single CNN generator | polygon segmentation · Cityscapes polygon segmentation (mean over 8 classes) | IoU: 71.38 -> 72.50 | Improved |
| replace Polygon-RNN++ with DDSL-based PolygonNet | polygon segmentation · Cityscapes polygon segmentation, batch of 16 crops | runtime (s): 2.3241 -> 0.0287 | Improved |
| replace Polygon-RNN++ with PolygonNet on per-class IoU | polygon segmentation · Cityscapes (Person) | IoU: 72.41 -> 68.62 | Mixed |
Closest papers by meaning
- Parallelised Differentiable Straightest Geodesics for 3D Meshes (CVPR 2026 · similarity 0.84)
- CvxNet: Learnable Convex Decomposition (CVPR 2020 · similarity 0.84)
- PoNQ: a Neural QEM-based Mesh Representation (CVPR 2024 · similarity 0.84)
- PUGeo-Net: A Geometry-centric Network for 3D Point Cloud Upsampling (ECCV 2020 · similarity 0.84)
- Flatten Anything: Unsupervised Neural Surface Parameterization (NeurIPS 2024 · similarity 0.83)
- Neural Processing of Tri-Plane Hybrid Neural Fields (ICLR 2024 · similarity 0.83)
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.
