Research map / 3D vision / Human pose estimation
TokenHPE: Learning Orientation Tokens for Efficient Head Pose Estimation via Transformers
Read the paperCheck 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 TokenHPE-v2 instead of 6DRepNet | HPE on BIWI with 70% train / 30% test · BIWI (70% train / 30% test) | MAE: 2.66 -> 2.49 | Helped |
| Use TokenHPE-v2 instead of TriNet | HPE on BIWI trained on 300W-LP · BIWI | MAEV: 5.86 -> 5.60 | Helped |
| Use TokenHPE-v2 instead of TriNet | HPE trained on 300W-LP · AFLW2000 | MAEV: 5.99 -> 5.98 | Helped |
| Replace no feature extractor with a ViT feature extractor (ablation) | TokenHPE · AFLW2000 | MAE: 5.38 -> 4.66 | Helped |
| Replace a CNN feature extractor with a ViT feature extractor (ablation) | TokenHPE · AFLW2000 | MAE: 4.96 -> 4.66 | Helped |
| Replace a CNN feature extractor with a ViT feature extractor (ablation) | TokenHPE · AFLW2000 | MAEV: 6.04 -> 5.98 | Helped |
| Add 2D sine positional embedding (ablation) | TokenHPE · AFLW2000 | MAE: 5.33 -> 4.66 | Helped |
| Use 2D sine instead of learnable positional embedding (ablation) | TokenHPE · AFLW2000 | MAE: 4.87 -> 4.66 | Helped |
| Increase orientation regions from nine to eleven (ablation) | TokenHPE · AFLW2000 | MAE: 4.85 -> 4.66 | Helped |
| Vary the weight α of prediction loss versus orientation-token loss (ablation) | TokenHPE · AFLW2000 | MAE | Mixed |
Closest papers by meaning
- Pose-Dependent Low-Rank Embedding for Head Pose Estimation (AAAI 2016 · similarity 0.86)
- Unsupervised Discovery of Facial Landmarks and Head Pose (CVPR 2025 · similarity 0.85)
- Self-Learning Transformations for Improving Gaze and Head Redirection (NeurIPS 2020 · similarity 0.85)
- SDPose: Tokenized Pose Estimation via Circulation-Guide Self-Distillation (CVPR 2024 · similarity 0.84)
- Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose Estimation (CVPR 2024 · similarity 0.83)
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.83)
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.