Research map / 3D vision / Human pose estimation
Learning Single-View 3D Reconstruction with Limited Pose Supervision
Guandao Yang, Yin Cui, Serge Belongie, Bharath Hariharan
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 unlabeled images to pose-annotated training (ablation) | single-category chair model · ShapeNet chairs (1% pose annotations) | IoU: 0.2913 -> 0.3175 | Improved |
| add unlabeled images to pose-annotated training (ablation) | single-category chair model · ShapeNet chairs (1% pose annotations) | AP: 0.3800 -> 0.4162 | Improved |
| add pose-annotated images to unlabeled-only training (ablation) | single-category chair model · ShapeNet chairs (1% pose annotations) | IoU: 0.2065 -> 0.3175 | Improved |
| pool multiple categories (category-agnostic) instead of chairs only (ablation) | pose+unlabeled chair model · ShapeNet chairs (1% pose annotations) | IoU: 0.3175 -> 0.3104 | Mixed |
| fine-tune the category-agnostic model on the target category (ablation) | category-agnostic model · ShapeNet chairs (1% pose annotations) | IoU: 0.3104 -> 0.3250 | Improved |
| use category-agnostic pooling at 50% pose supervision instead of category-specific (ablation) | Ours · ShapeNet 7 categories mean | IoU: 0.54 -> 0.55 | No clear effect |
| increase pose annotations from 50% to 100% in category-agnostic model (ablation) | Ours · ShapeNet 7 categories mean | AP: 0.70 -> 0.71 | No clear effect |
| increase pose annotations from 1% to 10% in category-agnostic model (ablation) | Ours · ShapeNet 7 categories mean | IoU: 0.42 -> 0.51 | Improved |
| add unlabeled images when pose supervision is 50% or 100% (ablation) | Ours · ShapeNet chairs | IoU | No clear effect |
| add unlabeled images when pose supervision is 10% or less (ablation) | Ours · ShapeNet chairs | IoU | Improved |
| fine-tune pre-trained category-agnostic model instead of training from scratch on novel category (ablation) | Ours · novel categories benches, vessels, cabinets (1% pose) | IoU and AP | Improved |
| fine-tune instead of applying pre-trained model out-of-category (ablation) | Ours · novel categories benches, vessels, cabinets (1% pose) | IoU and AP | Improved |
Closest papers by meaning
- Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose (NeurIPS 2021 · similarity 0.90)
- Shape-Pose Ambiguity in Learning 3D Reconstruction from Images (AAAI 2021 · similarity 0.89)
- Learning 3D Human Pose from Structure and Motion (ECCV 2018 · similarity 0.88)
- Weakly-Supervised 3D Human Pose Learning via Multi-View Images in the Wild (CVPR 2020 · similarity 0.88)
- Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · similarity 0.87)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · 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.
