Research map / 3D vision / Human pose estimation
SO-HandNet: Self-Organizing Network for 3D Hand Pose Estimation With Semi-Supervised Learning
Yujin Chen, Zhigang Tu, Liuhao Ge, Dejun Zhang, Ruizhi Chen, Junsong Yuan
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 unannotated data via point cloud autoencoder (semi-supervised) with 10% labels (ablation) | SO-HandNet · HANDS 2017 Frame-based | mean joint error (mm): 40.03 mm -> 24.65 mm | Improved |
| add semi-supervised training with unannotated data when only 25% labels are available (ablation) | SO-HandNet · ICVL (25% annotated) | mean joint error: 13.2% | Improved |
| add semi-supervised training with unannotated data when 50% labels are available (ablation) | SO-HandNet · ICVL (50% annotated) | mean joint error: 11.7% | Improved |
| add semi-supervised training with unannotated data when 75% labels are available (ablation) | SO-HandNet · ICVL (75% annotated) | mean joint error: 0.7% | No clear effect |
| add node features to global feature in HPE input (v1 fusion) (ablation) | SO-HandNet HPE · ICVL | mean joint error | Improved |
| fuse global and node features in M channels (v1) instead of kN channels (v2) (ablation) | SO-HandNet HPE · ICVL | mean joint error | Improved |
| add point features to global+node fusion (ablation) | SO-HandNet HPE · ICVL | mean joint error | Hurt |
| replace Crossing Nets with SO-HandNet at 25% labels | semi-supervised hand pose · NYU (25% annotated) | mean joint error (mm): 16.1 -> 14.9 | Improved |
| replace Crossing Nets with SO-HandNet at 25% labels | semi-supervised hand pose · ICVL (25% annotated) | mean joint error (mm): 10.5 -> 11.1 | Hurt |
| increase annotated data from 25% to 100% for SO-HandNet (ablation) | SO-HandNet · NYU | mean joint error (mm): 14.9 -> 11.2 | Improved |
| increase annotated data from 25% to 100% for Crossing Nets (ablation) | Crossing Nets · ICVL | mean joint error (mm): 10.5 -> 10.2 | No clear effect |
| use SO-HandNet instead of V2V-PoseNet for efficiency | hand pose estimator · ICVL (test, single GPU) | parameters / speed: 457.5M -> 16.6M; 3.5fps -> 58fps | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Crossing Nets — Crossing Nets: Combining GANs and VAEs With a Shared Latent Space for Hand Pose Estimation (CVPR 2017) · better on 7 of 8 shared results
- V2V-PoseNet — V2V-PoseNet: Voxel-to-Voxel Prediction Network for Accurate 3D Hand and Human Pose Estimation From a Single Depth Map (CVPR 2018) · better on 2 of 3 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
- Hand PointNet: 3D Hand Pose Estimation Using Point Sets (CVPR 2018 · similarity 0.93)
- Point-to-Point Regression PointNet for 3D Hand Pose Estimation (ECCV 2018 · similarity 0.92)
- Self-Supervised 3D Hand Pose Estimation Through Training by Fitting (CVPR 2019 · similarity 0.91)
- Two Heads Are Better than One: Image-Point Cloud Network for Depth-Based 3D Hand Pose Estimation (AAAI 2023 · similarity 0.91)
- Real-Time 3D Hand Pose Estimation with 3D Convolutional Neural Networks (TPAMI 2018 · similarity 0.91)
- HandFoldingNet: A 3D Hand Pose Estimation Network Using Multiscale-Feature Guided Folding of a 2D Hand Skeleton (ICCV 2021 · similarity 0.90)
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.
