Research map / Point clouds / Shape and surface completion
Learning 3D Shape Completion Under Weak Supervision
David Stutz, Andreas Geiger
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 |
|---|---|---|---|
| amortize maximum likelihood inference with a learned encoder (ML -> AML) | shape completion · SN-clean low res | IoU: 0.756 -> 0.779 | Improved |
| use probabilistic encoder with KL (AML) instead of deterministic dAML (ablation) | AML · SN-clean low res | IoU: 0.784 -> 0.779 | No clear effect |
| fuse multiple views (k=1 -> k=3) (ablation) | AML · SN-clean low res | IoU: 0.779 -> 0.809 | Improved |
| fuse k=5 views instead of k=3 (ablation) | AML · SN-clean low res | IoU: 0.809 -> 0.804 | No clear effect |
| fuse k=5 views with full supervision (single view -> k=5) (ablation) | Dai17 · SN-clean low res | IoU: 0.872 -> 0.924 | Improved |
| use skip connections (Sup -> Dai17) | fully supervised baseline · SN-clean low res | IoU: 0.841 -> 0.872 | Improved |
| train weakly supervised AML versus fully supervised Dai17 (SN-noisy) | AML · SN-noisy low res | IoU: 0.836 vs 0.771 | Mixed |
| increase resolution from low to medium | AML · SN-clean | IoU: 0.779 -> 0.788 | Improved |
| increase resolution from low to high | AML · SN-clean | IoU: 0.779 -> 0.796 | Improved |
| use AML versus data-driven Eng16 on KITTI | AML · KITTI low res | Completeness (m): 0.13 -> 0.12 | Improved |
| use full supervision Sup instead of AML | Sup · KITTI low res | Completeness (m): 0.12 -> 0.091 | Improved |
| train a single model across all ten ModelNet categories (AML vs Dai17 gap) | AML · ModelNet10 | IoU: 0.646 vs 0.509 | Mixed |
Closest papers by meaning
- Learning 3D Shape Completion From Laser Scan Data With Weak Supervision (CVPR 2018 · similarity 0.97)
- ShapeFormer: Transformer-Based Shape Completion via Sparse Representation (CVPR 2022 · similarity 0.90)
- 3D Shape Completion with Multi-View Consistent Inference (AAAI 2020 · similarity 0.90)
- Self-supervised Shape Completion via Involution and Implicit Correspondences (ECCV 2024 · similarity 0.90)
- FSC: Few-point Shape Completion (CVPR 2024 · similarity 0.89)
- Robust 3D Tracking with Quality-Aware Shape Completion (AAAI 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.
