Research map / 3D vision / Face and shape
TextureNet: Consistent Local Parametrizations for Learning From High-Resolution Signals on Meshes
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 TextureNet with high-resolution texture patches | 3D semantic segmentation · ScanNet v2 | mean class IoU: 48.4% vs. 56.6% | Helped |
| Use TextureNet with high-resolution texture patches | 3D semantic segmentation · Matterport3D | mean class accuracy: 56.1% vs. 63.0% | Helped |
| Use extrinsic rather than intrinsic 4-RoSy energy (ablation) | TextureNet · ScanNet v2 | mean class IoU: 44.4 -> 54.8 | Helped |
| Use extrinsic 4-RoSy directions rather than random local directions (ablation) | TextureNet · ScanNet v2 | mean class IoU: 41.1 -> 54.8 | Helped |
| Use extrinsic 4-RoSy directions rather than principal-curvature eigenvectors (ablation) | TextureNet · ScanNet v2 | mean class IoU: 42.5 -> 54.8 | Helped |
| Add per-point normal and RGB inputs (ablation) | TextureNet · ScanNet v2 | mean class IoU: 50.6 -> 54.8 | Helped |
| Replace per-point color with 10x10 high-resolution texture patches (ablation) | TextureNet · ScanNet v2 | mean class IoU: 54.8 -> 58.1; improves the mean class IoU by 3.3% | Helped |
| Use max pooling rather than average pooling (ablation) | TextureNet convolution aggregation · ScanNet v2, geometry-only | mean class IoU: 38.0 -> 50.6 | Helped |
| Use max pooling rather than average pooling (ablation) | TextureNet convolution aggregation · ScanNet v2, per-point normal and RGB | mean class IoU: 39.1 -> 54.8 | Helped |
| Canonicalize the 4-RoSy ambiguity rather than select an arbitrary cross direction (ablation) | Surface convolution · ScanNet v2, per-point normal and RGB | mean class IoU: 34.5 -> 54.8 | Helped |
| Use geodesic patches rather than Euclidean-ball neighborhoods (ablation) | PointNet++ · ScanNet v2 | mean class IoU: 35.0 -> 47.2 | Helped |
| Use QuadriFlow sampling rather than furthest point sampling (ablation) | PointNet++ · ScanNet v2 | mean class IoU: 51.6 -> 54.8 | Helped |
Closest papers by meaning
- Cross-Atlas Convolution for Parameterization Invariant Learning on Textured Mesh Surface (CVPR 2019 · similarity 0.91)
- Convolutional Generation of Textured 3D Meshes (NeurIPS 2020 · similarity 0.89)
- PFCNN: Convolutional Neural Networks on 3D Surfaces Using Parallel Frames (CVPR 2020 · similarity 0.88)
- AUV-Net: Learning Aligned UV Maps for Texture Transfer and Synthesis (CVPR 2022 · similarity 0.87)
- Flatten Anything: Unsupervised Neural Surface Parameterization (NeurIPS 2024 · similarity 0.87)
- TriTex: Learning Texture from a Single Mesh via Triplane Semantic Features (CVPR 2025 · 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.