Research map / 3D vision / Face and shape
Soft Rasterizer: A Differentiable Renderer for Image-Based 3D Reasoning
Shichen Liu, Tianye Li, Weikai Chen, Hao Li
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 Laplacian regularizer (ablation) | SoftRas · ShapeNet single-view reconstruction | mIoU: 62.4 v.s. 62.0 | Improved |
| add color loss (A_S color aggregation) (ablation) | SoftRas · ShapeNet single-view reconstruction | mIoU: 64.6 v.s. 62.4 | Improved |
| use Euclidean instead of barycentric distance (ablation) | SoftRas · ShapeNet single-view reconstruction | mIoU: 60.8 -> 62.0 | Improved |
| use neural-network aggregation A_N instead of A_O (ablation) | SoftRas · ShapeNet single-view reconstruction | mIoU: 62.4 -> 63.2 | Improved |
| add shading supervision to silhouette loss (ablation) | SoftRas · ShapeNet 13 categories | mean IoU: 0.6234 -> 0.6464 | Improved |
| replace NMR approximate-gradient renderer with SoftRas | mesh generator · ShapeNet 13 categories | mean IoU: 0.6015 -> 0.6464 | Improved |
| replace NMR with SoftRas | cube pose fitting · Cube rotation estimation (w/o scheduling) | mean relative angular error: 93.40 -> 82.80 | Improved |
| add optimization scheduling (decay sigma and gamma) (ablation) | SoftRas · Cube rotation estimation | mean relative angular error: 82.80 -> 63.57 | Improved |
| add multi-resolution scheduling (ablation) | NMR · Cube rotation estimation | mean relative angular error: 93.40 -> 80.94 | Improved |
Closest papers by meaning
- A General Differentiable Mesh Renderer for Image-Based 3D Reasoning (TPAMI 2020 · similarity 0.98)
- DRWR: A Differentiable Renderer without Rendering for Unsupervised 3D Structure Learning from Silhouette Images (ICML 2020 · similarity 0.91)
- Differentiable rendering with perturbed optimizers (NeurIPS 2021 · similarity 0.91)
- BG-Triangle: Bezier Gaussian Triangle for 3D Vectorization and Rendering (CVPR 2025 · similarity 0.88)
- LinPrim: Linear Primitives for Differentiable Volumetric Rendering (NeurIPS 2025 · similarity 0.88)
- UniGS: Unified Geometry-Aware Gaussian Splatting for Multimodal Rendering (NeurIPS 2026 · similarity 0.88)
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.
