Research map / 3D vision / Face and shape
Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction
Rohan Chabra, Jan E. Lenssen, Eddy Ilg, Tanner Schmidt, Julian Straub, Steven Lovegrove, Richard Newcombe
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 |
|---|---|---|---|
| use local latent codes per voxel instead of one global DeepSDF code | DeepLS · 3D Warehouse chair | Chamfer distance: 0.204 -> 0.030 | Improved |
| use local latent codes per voxel instead of one global DeepSDF code | DeepLS · 3D Warehouse lamp | Chamfer distance: 0.832 -> 0.078 | Improved |
| use local shapes with a much smaller decoder than DeepSDF | DeepLS · Stanford Bunny | training time to equal accuracy: one minute vs over 8 days of GPU time | Improved |
| replace TSDF Fusion with DeepLS on synthetic scenes | DeepLS · ICL-NUIM mean | surface error: 5.42 mm -> 4.92 mm | Improved |
| replace TSDF Fusion with DeepLS on real scans | DeepLS · 3D Scene Dataset Burghers | surface error: 10.11 -> 5.74 | Improved |
| replace TSDF Fusion with DeepLS on real scans | DeepLS · 3D Scene Dataset StoneWall | completion: 91.02 -> 91.37 | No clear effect |
| replace TSDF Fusion with DeepLS under artificial noise | DeepLS · ICL-NUIM with noise sigma 0.015 | surface error: 7.29 mm -> 6.41 mm | Improved |
| replace TSDF Fusion with DeepLS under artificial noise | DeepLS · ICL-NUIM with noise sigma 0.015 | completion: 68.53 % -> 71.04 % | Improved |
| increase compression (larger voxels) for DeepLS versus TSDF Fusion | DeepLS · ICL-NUIM representation size sweep | completeness: TSDF Fusion drops to about 50% less complete; DeepLS maintains completeness | Improved |
| use trilinear interpolation (partition of unity) between neighboring voxels (ablation) | DeepLS · decoder evaluation cost | decoder evaluations per query: factor of eight | Hurt |
| train codes with extended receptive field (L-infinity radius 1.5 voxel) for border consistency (ablation) | DeepLS · voxel boundary SDF | accuracy vs efficiency | Improved |
| enlarge receptive field of local shapes beyond a critical point (ablation) | DeepLS · 2D border consistency | SDF prediction error | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepSDF — DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (CVPR 2019) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Local Deep Implicit Functions for 3D Shape (CVPR 2020 · similarity 0.93)
- DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (CVPR 2019 · similarity 0.91)
- LP-DIF: Learning Local Pattern-Specific Deep Implicit Function for 3D Objects and Scenes (CVPR 2023 · similarity 0.90)
- High-Resolution Shape Completion Using Deep Neural Networks for Global Structure and Local Geometry Inference (ICCV 2017 · similarity 0.89)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.89)
- CoFie: Learning Compact Neural Surface Representations with Coordinate Fields (NeurIPS 2024 · 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.
