Research map / 3D vision / Human pose estimation
NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation
Angtian Wang, Adam Kortylewski, Alan Yuille
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 |
|---|---|---|---|
| Use feature-level NeMo rendering with occlusion/outlier inference instead of StarMap | 3D pose estimation · PASCAL3D+ L3 | ACC π/6 (%): 22.9 -> 47.1 | Improved |
| Use feature-level NeMo rendering with occlusion/outlier inference instead of StarMap | 3D pose estimation · PASCAL3D+ L2 | ACC π/6 (%): 47.2 -> 65.2 | Improved |
| Use feature-level NeMo rendering with occlusion/outlier inference instead of StarMap | 3D pose estimation · PASCAL3D+ L0 | ACC π/6 (%): 89.4 -> 86.7 | Hurt |
| Replace detailed mesh with category-specific cuboid meshes | NeMo · PASCAL3D+ L0 | ACC π/6 (%): 84.1 -> 86.7 | Improved |
| Replace detailed mesh with one category-level cuboid | NeMo · PASCAL3D+ L2 | ACC π/6 (%): 59.9 -> 63.9 | Improved |
| Remove outlier/clutter handling during inference (ablation) | NeMo-MultiCuboid · PASCAL3D+ L3 | ACC π/6 (%): 47.1 -> 44.4 | Hurt |
| Add outlier/clutter handling during inference (ablation) | NeMo-MultiCuboid · PASCAL3D+ L3 | Median error (degree): 41.6 -> 36.1 | Improved |
| Remove contrastive feature loss (ablation) | NeMo-MultiCuboid · PASCAL3D+ L3 | ACC π/6 (%): 47.1 -> 26.9 | Hurt |
| Add contrastive feature loss (ablation) | NeMo-MultiCuboid · PASCAL3D+ L3 | ACC π/18 (%): 5.6 -> 17.8 | Improved |
| Reduce random pose initialization samples from 144 to 72 (ablation) | NeMo-MultiCuboid · PASCAL3D+ | ACC π/6 (%): 86.7 -> 86.3 | No clear effect |
| Reduce random pose initialization samples from 144 to 1 (ablation) | NeMo-MultiCuboid · PASCAL3D+ | ACC π/6 (%): 86.7 -> 54.9 | Hurt |
| Use category-specific subtype information at inference (ablation) | NeMo-MultiCuboid · PASCAL3D+ L3 | ACC π/6 (%): 47.2 -> 47.1 | No clear effect |
Closest papers by meaning
- Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose (NeurIPS 2021 · similarity 0.88)
- Scaling 3D Compositional Models for Robust Classification and Pose Estimation (ICCV 2025 · similarity 0.87)
- Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · similarity 0.87)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.87)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.87)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · 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.
