Research map / 3D vision / Face and shape
Curriculum DeepSDF
Yueqi Duan, Haidong Zhu, He Wang, Li Yi Ram Nevatia, Leonidas J. Guibas
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 surface-accuracy curriculum (tolerance epsilon, fixed deepest network) to DeepSDF (ablation) | DeepSDF · ShapeNet (5 categories, average) | Chamfer distance, mean: 0.319 -> 0.257 | Improved |
| add progressive network growth to the surface-accuracy curriculum (ablation) | Ours-Sur · ShapeNet (5 categories, average) | Chamfer distance, mean: 0.257 -> 0.239 | Improved |
| add sample-difficulty curriculum to DeepSDF (ablation) | DeepSDF · ShapeNet (5 categories, average) | Chamfer distance, mean: 0.319 -> 0.246 | Improved |
| combine both curricula instead of surface accuracy only (ablation) | Ours-Sur · ShapeNet (5 categories, average) | Chamfer distance, mean: 0.239 -> 0.216 | Improved |
| combine both curricula instead of sample difficulty only (ablation) | Ours-Sam · ShapeNet (5 categories, average) | Chamfer distance, mean: 0.246 -> 0.216 | Improved |
| add sample-difficulty curriculum (lambda weighting) to the surface-accuracy curriculum (ablation) | Ours-Sur · ShapeNet lamp | Chamfer distance, mean: 0.639 -> 0.473 | Improved |
| gradually increase lambda instead of a fixed lambda 0.50 (ablation) | hard-sample weighting · ShapeNet lamp | Chamfer distance, mean: 0.508 -> 0.473 | Improved |
| raise fixed lambda from 0.50 to 0.75 (ablation) | hard-sample weighting · ShapeNet lamp | Chamfer distance, mean: 0.508 -> 0.567 | Hurt |
| increase only hard-sample weights (no semi-hard, no easy decrease) (ablation) | hard-sample weighting · ShapeNet lamp | Chamfer distance, mean: 0.639 -> 0.676 | Hurt |
| add semi-hard weight increase to hard-only weighting (ablation) | hard-sample weighting · ShapeNet lamp | Chamfer distance, mean: 0.676 -> 0.512 | Improved |
| add easy-sample weight decrease to hard+semi-hard weighting (ablation) | hard-sample weighting · ShapeNet lamp | Chamfer distance, mean: 0.512 -> 0.473 | Improved |
| apply Curriculum DeepSDF instead of DeepSDF for missing part recovery | DeepSDF · ShapeNet lamp, missing-part recovery (25% removed) | Chamfer distance: 4.79 -> 4.07 | Improved |
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 6 of 6 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
- DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (CVPR 2019 · similarity 0.91)
- Neural Geometric Level of Detail: Real-Time Rendering With Implicit 3D Shapes (CVPR 2021 · similarity 0.86)
- Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction (ECCV 2020 · similarity 0.86)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.86)
- Few-Shot Unsupervised Implicit Neural Shape Representation Learning with Spatial Adversaries (ICML 2024 · similarity 0.85)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.85)
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.
