Research map / 3D vision / Human pose estimation
HandDAGT: A Denoising Adaptive Graph Transformer for 3D Hand Pose Estimation
WENCAN CHENG, Eunji Kim, Jong Hwan Ko
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 |
|---|---|---|---|
| replace no attention with conventional dot-product softmax attention (ablation) | HandDAGT decoder · NYU | mean keypoint error (mm): 7.43 -> 10.66 | Hurt |
| use sigmoid Hadamard attention instead of conventional softmax attention (ablation) | HandDAGT decoder · NYU | mean keypoint error (mm): 10.38 -> 7.12 | Improved |
| add enhanced queries and keys (ablation) | HandDAGT decoder · NYU | mean keypoint error (mm): 7.30 -> 7.12 | Improved |
| add kinematic topology attention to local-point attention (ablation) | HandDAGT decoder · NYU | mean keypoint error (mm): 7.24 -> 7.12 | Improved |
| add local-point attention to kinematic topology attention (ablation) | HandDAGT decoder · NYU | mean keypoint error (mm): 7.21 -> 7.12 | Improved |
| add denoising training (ablation) | HandDAGT · NYU | mean keypoint error (mm): 7.20 -> 7.12 | Improved |
| stack 3 transformers instead of 1 (ablation) | HandDAGT · NYU | mean keypoint error (mm): 8.02 -> 7.12 | Improved |
| stack 3 transformers instead of 2 (ablation) | HandDAGT · NYU | mean keypoint error (mm): 7.15 -> 7.12 | No clear effect |
| increase input patch size from 64 to 128 super points (ablation) | HandDAGT · NYU | mean keypoint error (mm): 7.12 -> 7.16 | Hurt |
| reduce input patch size from 64 to 32 super points (ablation) | HandDAGT · NYU | mean keypoint error (mm): 7.12 -> 7.20 | Hurt |
| use HandDAGT instead of IPNet on HO3D | hand-object pose estimation · HO3D_v2 | mean keypoint error (mm): 1.81 -> 1.81 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports HandDAGT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- HandMCM: Multi-modal Point Cloud-based Correspondence State Space Model for 3D Hand Pose Estimation (AAAI 2026 · better on 7 of 7 shared results)
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.
- HandVoxNet++ — HandVoxNet++: 3D Hand Shape and Pose Estimation Using Voxel-Based Neural Networks (TPAMI 2021) · 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
- Deformer: Dynamic Fusion Transformer for Robust Hand Pose Estimation (ICCV 2023 · similarity 0.89)
- 3D Hand Pose Estimation via Articulated Anchor-to-Joint 3D Local Regressors (TPAMI 2025 · similarity 0.88)
- 3D Interacting Hand Pose Estimation by Hand De-Occlusion and Removal (ECCV 2022 · similarity 0.88)
- ManiDext: Hand-Object Manipulation Synthesis via Continuous Correspondence Embeddings and Residual-Guided Diffusion (TPAMI 2025 · similarity 0.88)
- HandDiff: 3D Hand Pose Estimation with Diffusion on Image-Point Cloud (CVPR 2024 · similarity 0.87)
- Physics-Aware Hand-Object Interaction Denoising (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.
