Research map / 3D vision / Human pose estimation
KeyPose: Multi-View 3D Labeling and Keypoint Estimation for Transparent Objects
Xingyu Liu, Rico Jonschkowski, Anelia Angelova, Kurt Konolige
Check 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 stereo RGB instead of monocular input (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 10.0 -> 4.6 | Improved |
| Use stereo RGB instead of monocular input (ablation) | KeyPose · TOD bottles | 3D MAE (mm): 10.1 -> 5.8 | Improved |
| Fuse the stereo views early rather than late (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 7.9 -> 4.6 | Improved |
| Fuse the stereo views early rather than late (ablation) | KeyPose · TOD bottles | 3D MAE (mm): 10.6 -> 5.8 | Improved |
| Add projection loss (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 5.4 -> 4.6 | Improved |
| Add projection loss (ablation) | KeyPose · TOD bottles | 3D MAE (mm): 9.9 -> 5.8 | Improved |
| Use direct regression instead of integral projection (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 4.7 -> 4.6 | Improved |
| Use direct regression instead of integral projection (ablation) | KeyPose · TOD bottles | 3D MAE (mm): 6.0 -> 5.8 | Improved |
| Add permutation loss for symmetric keypoints (ablation) | KeyPose · TOD tree_0 | 3D MAE (mm): 26.4 -> 12.8 | Improved |
| Increase crop size to 270x180 (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 4.6 -> 5.0 | Hurt |
| Increase crop size to 360x240 (ablation) | KeyPose · TOD bottle_0 | 3D MAE (mm): 4.6 -> 5.3 | Hurt |
| Use stereo RGB keypoint estimation rather than DenseFusion with opaque depth | KeyPose vs DenseFusion · TOD instance-level, held-out textures (mean) | MAE: 35.1 -> 9.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports KeyPose as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- "Category-level Object Detection, Pose Estimation and Reconstruction from Stereo Images" (ECCV 2024 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DenseFusion — DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion (CVPR 2019) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning (NeurIPS 2018 · similarity 0.89)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.88)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (CVPR 2024 · similarity 0.88)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.88)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.87)
- KeyPose: Category-Level 6D Object Pose Estimation with Self-Adaptive Keypoints (AAAI 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.
