Research map / 3D vision / 3D generation
Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision
Read the paperCheck 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 |
|---|---|---|---|
| add projection (silhouette) loss without volume supervision (PTN-Proj vs CNN-Vol) | single-category chair reconstruction · ShapeNetCore chair test | IU: 0.5027 vs 0.4983 | No clear effect |
| add volume loss to projection loss (PTN-Comb vs PTN-Proj) | single-category chair reconstruction · ShapeNetCore chair test | IU: 0.5067 vs 0.5027 | No clear effect |
| combined loss vs volume loss only | multi-category reconstruction · ShapeNetCore 13-category test | IU: best in 8 of 13 categories | Helped |
| projection loss only with partial views (chair-N) vs volume loss only (ablation) | single-category chair with narrow azimuth range · ShapeNetCore chair-N test | IU: 0.4583 vs 0.4380 (CNN-Vol) | Helped |
| add projection loss for out-of-category generalization (PTN-Proj vs CNN-Vol) | single-category training, unseen categories · out-of-category tests (bed, bookshelf, cabinet, motorbike, train) | IU: best in all 5 categories, e.g. cabinet 0.3937 vs 0.2588 | Helped |
| projection loss only vs combined/volume loss, multi-category training | out-of-category generalization · out-of-category tests, multi-category training | IU: PTN-Proj better on bed, bookshelf, cabinet, motorbike; PTN-Comb better on train (0.3744 vs 0.3670) | Mixed |
| projection loss only vs combined loss on multi-category test | multi-category reconstruction · ShapeNetCore 13-category test | IU: comparable; e.g. airplane 0.5556 vs 0.5836 | No clear effect |
| multi-category training vs single-category training | out-of-category generalization · out-of-category tests | IU: PTN-Proj multi bed 0.1944 vs single 0.1801; cabinet 0.6484 vs 0.3937 | Helped |
| train on partial azimuth views (chair-N) vs full views (ablation) | PTN-Comb single-category · ShapeNetCore chair test | IU: 0.4429 vs 0.5067 | Hurt |
Closest papers by meaning
- What Do Single-View 3D Reconstruction Networks Learn? (CVPR 2019 · similarity 0.91)
- Learning View Priors for Single-View 3D Reconstruction (CVPR 2019 · similarity 0.89)
- Higher-Order Function Networks for Learning Composable 3D Object Representations (ICLR 2020 · similarity 0.88)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.87)
- Triplane Meets Gaussian Splatting: Fast and Generalizable Single-View 3D Reconstruction with Transformers (CVPR 2024 · similarity 0.86)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · 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.