Research map / 3D vision / Face and shape
DRWR: A Differentiable Renderer without Rendering for Unsupervised 3D Structure Learning from Silhouette Images
Zhizhong Han, Chao Chen, Yu-Shen Liu, Matthias Zwicker
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 |
|---|---|---|---|
| replace rendering-based pixel losses with projection-based DRWR losses | unsupervised ShapeNet point-cloud reconstruction · ShapeNet mean, 32² images / 2000 points | CD: 4.61 -> 4.16 | Improved |
| replace rendering-based pixel losses with projection-based DRWR losses | unsupervised ShapeNet point-cloud reconstruction · ShapeNet mean, 64² images / 8000 points | CD: 3.55 -> 3.36 | Improved |
| replace rendering-based pixel losses with projection-based DRWR losses | unsupervised ShapeNet point-cloud reconstruction · ShapeNet mean, 128² images / 16000 points | CD: 2.96 -> 2.85 | Improved |
| use DRWR rather than the strongest unsupervised renderer | ShapeNet Cars · ShapeNet Cars | IoU: 78.8 -> 75.3 | Hurt |
| use DRWR rather than the strongest unsupervised renderer | ShapeNet Airplanes · ShapeNet Airplanes | IoU: 57.0 -> 62.2 | Improved |
| use DRWR rather than the strongest unsupervised renderer | ShapeNet Chairs · ShapeNet Chairs | IoU: 52.7 -> 58.1 | Improved |
| remove the structure-aware repulsion loss (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.01 -> 19.50 | Hurt |
| remove the smooth silhouette loss (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.01 -> 139.10 | Hurt |
| replace the smooth silhouette loss with binary pixel loss (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.01 -> 24.59 | Hurt |
| add indicator weights to the repulsion loss (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.58 -> 4.01 | Improved |
| add boundary bias to the repulsion loss (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.41 -> 4.01 | Improved |
| train with 4 views instead of 2 (ablation) | DRWR · ShapeNet Airplanes, 2000 points | CD: 4.79 -> 4.01 | Improved |
Closest papers by meaning
- Soft Rasterizer: A Differentiable Renderer for Image-Based 3D Reasoning (ICCV 2019 · similarity 0.91)
- A General Differentiable Mesh Renderer for Image-Based 3D Reasoning (TPAMI 2020 · similarity 0.90)
- GenDR: A Generalized Differentiable Renderer (CVPR 2022 · similarity 0.88)
- Unsupervised Learning of Fine Structure Generation for 3D Point Clouds by 2D Projections Matching (ICCV 2021 · similarity 0.87)
- RenderNet: A deep convolutional network for differentiable rendering from 3D shapes (NeurIPS 2018 · similarity 0.87)
- Robust Geometry-Preserving Depth Estimation Using Differentiable Rendering (ICCV 2023 · 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.
