Research map / 3D vision / Face and shape
Multiview Aggregation for Learning Category-Specific Shape Reconstruction
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 |
|---|---|---|---|
| Represent visible and occluded surfaces with X-NOCS maps rather than directly predicting a point cloud | Single-view reconstruction · ShapeNetPlain, Cars; Airplanes; Chairs | Chamfer distance: Cars: 0.2932 -> 0.1569; Airplanes: 0.2549 -> 0.1855; Chairs: 0.4314 -> 0.3803 | Helped |
| Use X-NOCS maps and multiview aggregation | 3D reconstruction · ShapeNetPlain, Chairs, 3 and 5 views | Chamfer distance: 3 views: 0.2191 -> 0.1952; 5 views: 0.1932 -> 0.1576 | Helped |
| Use X-NOCS maps and multiview aggregation | 3D reconstruction · ShapeNetPlain, Chairs, 2 views | Chamfer distance: 0.2511 -> 0.2508 | No clear effect |
| Add mask loss and mask prediction to L2 shape prediction (ablation) | Single-view reconstruction · ShapeNetCOCO, Cars; Airplanes; Chairs | Chamfer distance: Cars: 3.6573 -> 0.5093; Airplanes: 7.9072 -> 0.3037; Chairs: 4.4716 -> 0.4401 | Helped |
| Add peeled-color prediction (ablation) | L2+Mask single-view model · ShapeNetCOCO, Cars; Airplanes; Chairs | Chamfer distance: Cars: 0.5093 -> 0.3714; Airplanes: 0.3037 -> 0.2659; Chairs: 0.4401 -> 0.4288 | Helped |
| Aggregate independent-view shape predictions using set union (ablation) | Single-view model · ShapeNetCOCO, Cars; Airplanes; Chairs, 2–5 views | Chamfer distance: Cars: 0.4206 -> 0.3692 (2 views); 0.3974 -> 0.3537 (3); 0.3692 -> 0.2731 (5). Chairs: 0.4249 -> 0.3649 (2); 0.3813 -> 0. | Helped |
| Aggregate independent-view shape predictions using set union (ablation) | Single-view model · ShapeNetCOCO, Airplanes, 2–3 views | Chamfer distance: 2 views: 0.1760 -> 0.2387; 3 views: 0.1677 -> 0.1782 | Hurt |
| Add permutation-equivariant feature-level aggregation (ablation) | Multiview shape reconstruction · ShapeNetCOCO, Cars; Chairs, 2–5 views | Chamfer distance: Cars: 0.4206 -> 0.3789 (2 views); 0.3974 -> 0.3537 (3); 0.3692 -> 0.2731 (5). Chairs: 0.4249 -> 0.3649 (2); 0.3813 -> 0. | Helped |
| Add permutation-equivariant feature-level aggregation (ablation) | Multiview shape reconstruction · ShapeNetCOCO, Airplanes, 2–3 views | Chamfer distance: 2 views: 0.1760 -> 0.2387; 3 views: 0.1677 -> 0.1782 | Hurt |
| Add more input views to the multiview model (ablation) | Permutation-equivariant multiview model · ShapeNetCOCO, Cars; Airplanes; Chairs | Chamfer distance: Cars: 0.3789 -> 0.2731; Airplanes: 0.2387 -> 0.1277; Chairs: 0.3649 -> 0.2457 | Helped |
| Train with a variable number of views up to 5 instead of a fixed number (ablation) | Multiview model · ShapeNetPlain, Cars; Airplanes; Chairs, 2–5 views | Chamfer distance: Cars: 0.2645 vs 0.2896 (2); 0.1645 vs 0.1989 (3); 0.1721 vs 0.1955 (5). Airplanes: 0.1318 vs 0.1418 (2); 0.1571 vs 0.100 | Mixed |
| Train with a variable number of views up to 10 instead of a fixed number (ablation) | Multiview model · ShapeNetPlain, Cars; Airplanes; Chairs, 2–5 views | Chamfer distance: Cars: 0.2645 vs 0.2992 (2); 0.1645 vs 0.2447 (3); 0.1721 vs 0.3095 (5). Airplanes: 0.1318 vs 0.1847 (2); 0.1571 vs 0.130 | Mixed |
Closest papers by meaning
- Learning to Reconstruct Shapes from Unseen Classes (NeurIPS 2018 · similarity 0.90)
- Learning View Priors for Single-View 3D Reconstruction (CVPR 2019 · similarity 0.89)
- Unsupervised Learning of 3D Object Categories From Videos in the Wild (CVPR 2021 · similarity 0.89)
- Pixels, Voxels, and Views: A Study of Shape Representations for Single View 3D Object Shape Prediction (CVPR 2018 · similarity 0.89)
- Multiview Compressive Coding for 3D Reconstruction (CVPR 2023 · similarity 0.89)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.87)
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.