Research map / 3D vision / Human pose estimation
Face-from-Depth for Head Pose Estimation on Depth Images
Guido Borghi, Matteo Fabbri, Roberto Vezzani, Simone Calderara, Rita Cucchiara
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 head crop around ground-truth head position to depth-only network (ablation) | depth CNN · Pandora head pose | accuracy: 0.553 -> 0.646 | Improved |
| add FfD branch to depth branch via concatenation (ablation) | POSEidon+ · Pandora head pose | accuracy: 0.646 -> 0.698 | Improved |
| add Motion Images branch to depth branch via concatenation (ablation) | POSEidon+ · Pandora head pose | accuracy: 0.646 -> 0.690 | Improved |
| add Motion Images as third branch to depth+FfD with conv+concat fusion (ablation) | POSEidon+ · Pandora head pose | accuracy: 0.698 -> 0.736 | Improved |
| replace gray-level-trained branch with FfD-generated input branch (ablation) | POSEidon+ · Pandora head pose | accuracy: 0.720 -> 0.736 | Improved |
| use new det-cGAN FfD instead of preliminary FfD in the framework | POSEidon+ · Pandora head pose | accuracy: 0.715 -> 0.736 | Improved |
| add U-Net skip connections to the FfD generator (ablation) | FfD · Pandora face reconstruction | L1 norm: 18.21 -> 23.75 | Hurt |
| use det-cGAN FfD instead of pix2pix for reconstruction | FfD · Pandora head pose (gray-trained net) | accuracy: 0.581 -> 0.613 | Improved |
| test gray-trained pose net on FfD faces instead of real gray-level | head pose network · Pandora head pose (gray-trained net) | accuracy: 0.613 -> 0.613 | No clear effect |
| crop shoulder input with Rx=850, Ry=500 (ablation) | shoulder pose network · Pandora shoulder pose | accuracy: 0.877 -> 0.924 | Improved |
| use narrow shoulder crop Rx=700, Ry=250 (ablation) | shoulder pose network · Pandora shoulder pose | accuracy: 0.877 -> 0.845 | Hurt |
| POSEidon+ versus preliminary POSEidon | POSEidon+ · Biwi (fixed split) | avg angle error: 1.7 -> 1.6 | Improved |
Closest papers by meaning
- POSEidon: Face-From-Depth for Driver Pose Estimation (CVPR 2017 · similarity 0.93)
- Multi-Task Head Pose Estimation in-the-Wild (TPAMI 2020 · similarity 0.87)
- Unsupervised Depth Estimation, 3D Face Rotation and Replacement (NeurIPS 2018 · similarity 0.86)
- HeadFusion: 360 Head Pose Tracking Combining 3D Morphable Model and 3D Reconstruction (TPAMI 2018 · similarity 0.85)
- Head2Body: Body Pose Generation from Multi-sensory Head-mounted Inputs (ICCV 2025 · similarity 0.85)
- Deep 3D Portrait From a Single Image (CVPR 2020 · similarity 0.85)
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.
