Research map / 3D vision / Face and shape
Few-Shot Generalization for Single-Image 3D Reconstruction via Priors
Bram Wallace, Bharath Hariharan
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 a category-specific prior shape as input | image-to-voxel network · ShapeNet novel classes | mean IoU: 0.36 -> 0.39 | Improved |
| increase number of novel-class shapes averaged into the prior from 1 to 10 | 1-Iteration 1-Shot · ShapeNet novel classes | mean IoU: 0.38 -> 0.40 | Improved |
| use prior alone without the image (1-shot guess) (ablation) | 1-Iteration 1-Shot · ShapeNet novel classes | mean IoU: 0.39 -> 0.23 | Hurt |
| add iterative refinement (3 iterations) at test time on single view (ablation) | 1-Iteration 1-Shot · ShapeNet novel classes | mean IoU: 0.39 vs 0.37 | No clear effect |
| add iterative refinement with more views | 3-Iteration · ShapeNet novel classes multi-view | mean IoU: 0.34 -> 0.41 | Improved |
| add more views to non-iterative model | 1-iteration 1-shot · ShapeNet novel classes multi-view | mean IoU: 0.39 -> 0.39 | No clear effect |
| add more views (1 to 5) to 3-iteration model | 3-Iteration · ShapeNet base classes multi-view | mean IoU: 0.61 -> 0.64 | Improved |
| use a prior from a wrong category (ablation) | 1-Iteration models · ShapeNet novel classes | mean IoU: never reach baseline 0.36 | Hurt |
| add 3-iteration refinement on cabinets | 3-Iteration · Cabinets (novel) | mean IoU: 0.66 -> 0.62 | Hurt |
| add category prior on rifles | 1-Iteration 1-Shot · Rifles (novel) | mean IoU: 0.12 -> 0.19 | Improved |
| train with the full prior rather than no prior on base classes | 1 Iteration Full Prior · ShapeNet base classes | mean IoU: 0.62 -> 0.63 | No clear effect |
| use target shape as input prior (ablation) | 1-iteration 1-shot · ShapeNet novel classes | mean IoU: 0.39 -> 0.41 | Improved |
Closest papers by meaning
- Fostering Generalization in Single-View 3D Reconstruction by Learning a Hierarchy of Local and Global Shape Priors (CVPR 2021 · similarity 0.92)
- Few-Shot Single-View 3-D Object Reconstruction with Compositional Priors (ECCV 2020 · similarity 0.92)
- Learning View Priors for Single-View 3D Reconstruction (CVPR 2019 · similarity 0.91)
- Few-Shot Single-View 3D Reconstruction with Memory Prior Contrastive Network (ECCV 2022 · similarity 0.91)
- Pre-Train, Self-Train, Distill: A Simple Recipe for Supersizing 3D Reconstruction (CVPR 2022 · similarity 0.91)
- 3D Reconstruction with Generalizable Neural Fields using Scene Priors (ICLR 2024 · 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.
