Research map / 3D vision / Human pose estimation
HOPE-Net: A Graph-Based Model for Hand-Object Pose Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the Adaptive Graph U-Net instead of three fully connected layers (ablation) | HOPE-Net 2D-to-3D converter · 2D-to-3D hand and object pose estimation | Average error (mm): 185.18 -> 6.81 | Helped |
| Add pooling and unpooling to the adaptive graph convolution network (ablation) | HOPE-Net 2D-to-3D converter · 2D-to-3D hand and object pose estimation | Average error (mm): 68.93 -> 6.81 | Helped |
| Use trainable pooling instead of gPool (ablation) | Adaptive Graph U-Net · 2D-to-3D hand and object pose estimation | Average error (mm): 153.28 -> 6.81 | Helped |
| Use trainable pooling instead of fixed pooling (ablation) | Adaptive Graph U-Net · 2D-to-3D hand and object pose estimation | Average error (mm): 7.41 -> 6.81 | Helped |
| Initialize the adjacency matrix with identity rather than zeros (ablation) | Adaptive Graph U-Net · Adaptive graph convolution 3D pose estimation | Average error (mm): 92805.02 -> 6.81 | Helped |
| Initialize the adjacency matrix with identity rather than random connections (ablation) | Adaptive Graph U-Net · Adaptive graph convolution 3D pose estimation | Average error (mm): 94.42 -> 6.81 | Helped |
| Initialize the adjacency matrix with identity rather than all ones (ablation) | Adaptive Graph U-Net · Adaptive graph convolution 3D pose estimation | Average error (mm): 63.25 -> 6.81 | Helped |
| Initialize the adjacency matrix with identity rather than the hand skeleton and object bounding box (ablation) | Adaptive Graph U-Net · Adaptive graph convolution 3D pose estimation | Average error (mm): 12.91 -> 6.81 | Helped |
| Pretrain on synthetic ObMan images, then fine-tune on real images | HOPE-Net · FPHA and HO-3D | pose estimation accuracy | Helped |
| Use detection-based 2D keypoint prediction on FPHA, where many points fall outside the image | HOPE-Net image encoder · First-Person Hand Action Dataset | 2D keypoint estimation | Hurt |
Closest papers by meaning
- Collaborative Learning for Hand and Object Reconstruction With Attention-Guided Graph Convolution (CVPR 2022 · similarity 0.90)
- 3D Hand Shape and Pose Estimation From a Single RGB Image (CVPR 2019 · similarity 0.90)
- Hand PointNet: 3D Hand Pose Estimation Using Point Sets (CVPR 2018 · similarity 0.90)
- Depth-Based 3D Hand Pose Estimation: From Current Achievements to Future Goals (CVPR 2018 · similarity 0.89)
- Harmonious Feature Learning for Interactive Hand-Object Pose Estimation (CVPR 2023 · similarity 0.89)
- DeepHand: Robust Hand Pose Estimation by Completing a Matrix Imputed With Deep Features (CVPR 2016 · similarity 0.89)
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.