Research map / 3D vision / Human pose estimation
POSEidon: Face-From-Depth for Driver 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 |
|---|---|---|---|
| Add depth, Face-from-Depth, and motion-image streams, fused with convolution followed by concatenation (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.657 -> 0.715 | Helped |
| Add the Face-from-Depth stream to depth and motion-image inputs, using convolution-plus-concatenation fusion (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.690 -> 0.715 | Helped |
| Use convolution-plus-concatenation rather than concatenation alone for three-stream fusion (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.657 -> 0.715 | Helped |
| Use multiplication-plus-concatenation rather than concatenation alone for three-stream fusion (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.657 -> 0.712 | Helped |
| Add cropping around the head to the single depth CNN (ablation) | Single-CNN depth pose estimator · Pandora | head pose accuracy (err < 15°): 0.553 -> 0.646 | Helped |
| Use only Face-from-Depth instead of depth input (ablation) | Single-CNN pose estimator · Pandora | head pose accuracy (err < 15°): 0.646 -> 0.647 | No clear effect |
| Use depth and motion images without the Face-from-Depth stream (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.715 -> 0.690 | Hurt |
| Use only motion images rather than the full three-stream model (ablation) | POSEidon head pose network · Pandora | head pose accuracy (err < 15°): 0.715 -> 0.609 | Hurt |
| Use shoulder crop Rx=850, Ry=500 rather than no crop (ablation) | POSEidon shoulder network · Pandora | shoulder pose accuracy (err < 15°): 0.877 -> 0.924 | Helped |
| Use shoulder crop Rx=700, Ry=250 rather than no crop (ablation) | POSEidon shoulder network · Pandora | shoulder pose accuracy (err < 15°): 0.877 -> 0.845 | Hurt |
| Occlude the upper part of the head (ablation) | POSEidon · Biwi | head pitch error (degrees): 42.5 $\pm$ 21.2 | Hurt |
| Occlude the middle of the face around the nose (ablation) | POSEidon · Biwi | head pitch error (degrees): 11.0 $\pm$ 5.3 | Hurt |
Closest papers by meaning
- FIFA: Fine-grained Inter-frame Attention for Driver's Video Gaze Estimation (CVPR 2025 · similarity 0.84)
- TokenHPE: Learning Orientation Tokens for Efficient Head Pose Estimation via Transformers (CVPR 2023 · similarity 0.84)
- Unsupervised Depth Estimation, 3D Face Rotation and Replacement (NeurIPS 2018 · similarity 0.83)
- Neural Head Reenactment with Latent Pose Descriptors (CVPR 2020 · similarity 0.83)
- Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction (CVPR 2020 · similarity 0.83)
- What Do You See in Vehicle? Comprehensive Vision Solution for In-Vehicle Gaze Estimation (CVPR 2024 · similarity 0.82)
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.