Research map / 3D vision / Face and shape
Unsupervised 3D Reconstruction Networks
Geonho Cha, Minsik Lee, Songhwai Oh
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 low-rank prior loss (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.175 -> 0.168 | Improved |
| add weight regularization loss (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.174 -> 0.168 | Improved |
| add intermediate losses on every module (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.268 -> 0.168 | Improved |
| increase number of 3D reconstructors from 2 to 12 (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.257 -> 0.168 | Improved |
| increase reconstructors from 3 to 9 (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.211 -> 0.211 | No clear effect |
| use nuclear norm rather than determinant cost for low-rank loss (ablation) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: +0.009 error with determinant | Improved |
| use nuclear norm rather than log-determinant cost for low-rank loss (ablation) | 3D-URN · PASCAL3D+ | training stability: log-determinant fails | Improved |
| add noise to 2D points (0.01 dmax) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.168 -> 0.177 | Hurt |
| double the noise std | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.177 -> 0.189 | Hurt |
| evaluate on unseen instances (80/20 split) | 3D-URN · PASCAL3D+ | normalized mean 3D error e1_3D: 0.168 -> 0.166 | No clear effect |
| apply to non-rigid face sequence versus NRSfM methods | 3D-URN · Face sequence | e2_3D: 0.023 (BMM) vs 0.044 | Hurt |
Closest papers by meaning
- Toward Realistic Single-View 3D Object Reconstruction With Unsupervised Learning From Multiple Images (ICCV 2021 · similarity 0.88)
- 3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction (ECCV 2016 · similarity 0.87)
- PR-RRN: Pairwise-Regularized Residual-Recursive Networks for Non-Rigid Structure-From-Motion (ICCV 2021 · similarity 0.87)
- LASR: Learning Articulated Shape Reconstruction From a Monocular Video (CVPR 2021 · similarity 0.87)
- Unsupervised 3D Reconstruction and Grouping of Rigid and Non-Rigid Categories (TPAMI 2020 · similarity 0.86)
- Implicit Functions in Feature Space for 3D Shape Reconstruction and Completion (CVPR 2020 · 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.
