Research map / 3D vision / Face and shape
Representing 3D Faces with Learnable B-Spline Volumes
Prashanth Chandran, Daoye Wang, Timo Bolkart
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 residual MLP (ablation) | CUBE-L 16^3 · Synthetic scans test (30,000) | PTS mean (mm): 1.65 -> 1.40 | Improved |
| add residual MLP (ablation) | CUBE-L 4^3 · Synthetic scans test (30,000) | PTS mean (mm): 2.63 -> 1.62 | Improved |
| add residual MLP (ablation) | CUBE-L 8^3 · Synthetic scans test (30,000) | V2V mean (mm): 2.44 -> 1.50 | Improved |
| add residual MLP (ablation) | CUBE-S 4^3 · Synthetic scans test (30,000) | PTS mean (mm): 2.59 -> 1.77 | Improved |
| add residual MLP (ablation) | CUBE-S 16^3 · Synthetic scans test (30,000) | PTS mean (mm): 1.74 -> 1.64 | Improved |
| add residual MLP (ablation) | CUBE-L · Real multi-view scans (18,000) | PTS mean (mm): 1.12 -> 0.95 | Improved |
| increase control points from 4^3 to 16^3 (ablation) | CUBE-L with residual MLP · Synthetic scans test (30,000) | PTS mean (mm): 1.62 -> 1.40 | Improved |
| increase control points from 4^3 to 16^3 (ablation) | CUBE-L without residual MLP · Synthetic scans test (30,000) | PTS mean (mm): 2.63 -> 1.65 | Improved |
| increase encoder size from S to L (ablation) | CUBE 16^3 with residual MLP · Synthetic scans test (30,000) | PTS mean (mm): 1.64 -> 1.40 | Improved |
| increase encoder size from S to L (ablation) | CUBE 4^3 without residual MLP · Synthetic scans test (30,000) | PTS mean (mm): 2.59 -> 2.63 | No clear effect |
| use CUBE-L with residuals instead of TEMPEH with refinement | TEMPEH w/ Refinement · Real multi-view scans (18,000) | PTS mean (mm): 1.13 -> 0.95 | Improved |
| use CUBE-L with residuals instead of BPS | BPS 1024 points · Real multi-view scans (18,000) | V2V mean (mm): 15.16 -> 1.69 | Improved |
Closest papers by meaning
- Shape My Face: Registering 3D Face Scans by Surface-to-Surface Translation (IJCV 2021 · similarity 0.88)
- Generating 3D Faces using Convolutional Mesh Autoencoders (ECCV 2018 · similarity 0.87)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.87)
- Topologically Consistent Multi-View Face Inference Using Volumetric Sampling (ICCV 2021 · similarity 0.87)
- FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation (CVPR 2026 · similarity 0.86)
- Pixel3DMM: Versatile Screen-Space Priors for Single-Image 3D Face Reconstruction (ICLR 2026 · similarity 0.86)
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.
