Research map / 3D vision / Face and shape
Multi-View Consistency as Supervisory Signal for Learning Shape and Pose Prediction
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 |
|---|---|---|---|
| learn shape and pose jointly using multi-view consistency without pose supervision | single-view shape and pose prediction · ShapeNet mean | mean IoU: 0.62 vs 0.55 | Helped |
| use depth rather than mask verification (ablation) | multi-view training without rotation supervision · ShapeNet mean | mean IoU: 0.55 vs 0.53 | No clear effect |
| remove known camera translation (ablation) | multi-view training without rotation supervision (Mask) · ShapeNet mean | mean IoU: 0.55 -> 0.40 | Hurt |
| remove known camera translation (ablation) | multi-view training without rotation supervision (Depth) · ShapeNet mean | mean IoU: 0.53 -> 0.47 | Hurt |
| remove known camera translation (ablation) | multi-view pose prediction with mask verification · ShapeNet mean | pose Accπ/6: 0.79 -> 0.54 | Hurt |
| remove known camera translation (ablation) | multi-view pose prediction with depth verification · ShapeNet mean | pose Accπ/6: 0.76 -> 0.61 | Hurt |
| add an approximate pose prior and remove it midway through training (ablation) | pose prediction network · ShapeNet | training success | Helped |
| remove the pose prior (ablation) | pose prediction network · ShapeNet | pose diversity | Hurt |
| use depth instead of mask verification (ablation) | shape prediction for chairs · ShapeNet chairs | shape concavity recovery | Helped |
Closest papers by meaning
- Shape-Pose Ambiguity in Learning 3D Reconstruction from Images (AAAI 2021 · similarity 0.89)
- Unsupervised Learning of Shape and Pose with Differentiable Point Clouds (NeurIPS 2018 · similarity 0.88)
- Multi-View Supervision for Single-View Reconstruction via Differentiable Ray Consistency (CVPR 2017 · similarity 0.87)
- ShapeClipper: Scalable 3D Shape Learning From Single-View Images via Geometric and CLIP-Based Consistency (CVPR 2023 · similarity 0.87)
- Learning 3D Object Shape and Layout Without 3D Supervision (CVPR 2022 · similarity 0.86)
- Novel Object Viewpoint Estimation Through Reconstruction Alignment (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.