Research map / 3D vision / Human pose estimation
Self-Supervised 3D Hand Pose Estimation From Monocular RGB via Contrastive Learning
Adrian Spurr, Aneesh Dahiya, Xi Wang, Xucong Zhang, Otmar Hilliges
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 equivariant contrastive loss (PeCLR) instead of SimCLR NT-Xent (ablation) | RN50 frozen encoder · FreiHAND | 3D EPE: 16.62 -> 16.05 | Improved |
| use equivariant contrastive loss (PeCLR) instead of SimCLR NT-Xent (ablation) | RN50 frozen encoder · FreiHAND | 2D EPE: 12.05 -> 10.51 | Improved |
| use equivariant loss for translation augmentation (ablation) | RN50 frozen encoder · FreiHAND | 3D EPE: 34% relative improvement | Improved |
| use equivariant loss for rotation augmentation (ablation) | RN50 frozen encoder · FreiHAND | 3D EPE: 57% relative improvement | Improved |
| add PeCLR pre-training on FH and YT3D (ablation) | RN50 · FreiHAND test | 3D PA-EPE: 0.83 -> 0.71 | Improved |
| add PeCLR pre-training on FH and YT3D (ablation) | RN152 · FreiHAND test | 3D PA-EPE: 0.74 -> 0.66 | Improved |
| add PeCLR pre-training (ablation) | RN152 · YouTube3DHands cross-dataset | 2D EPE: 22.1 -> 16.9 | Improved |
| add PeCLR pre-training (ablation) | RN152 · YouTube3DHands cross-dataset | 3D PA-EPE: 3.05 -> 2.88 | Improved |
| add PeCLR pre-training (ablation) | RN152 · FreiHAND cross-dataset eval | 3D EPE: 5.05 -> 4.56 | Improved |
| add unlabeled YT3D data to PeCLR pre-training (FH+YT3D vs FH) (ablation) | RN152 · FreiHAND semi-supervised | 3D EPE | Improved |
| add PeCLR pre-training with limited labels (ablation) | RN152 · FreiHAND semi-supervised | 3D EPE | Improved |
| add rotation augmentation with SimCLR (invariant loss) (ablation) | SimCLR · FreiHAND | 3D EPE | Hurt |
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.
- I2L-MeshNet — I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image (ECCV 2020) · better on 1 of 1 shared result
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
- Cross-Domain 3D Hand Pose Estimation With Dual Modalities (CVPR 2023 · similarity 0.90)
- Aligning Latent Spaces for 3D Hand Pose Estimation (ICCV 2019 · similarity 0.89)
- Self-Supervised 3D Hand Pose Estimation Through Training by Fitting (CVPR 2019 · similarity 0.89)
- SiMHand: Mining Similar Hands for Large-Scale 3D Hand Pose Pre-training (ICLR 2025 · similarity 0.89)
- Model-Based 3D Hand Reconstruction via Self-Supervised Learning (CVPR 2021 · similarity 0.89)
- Keypoint Fusion for RGB-D Based 3D Hand Pose Estimation (AAAI 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.
