Research map / 3D vision / Human pose estimation
Towards Unbiased Label Distribution Learning for Facial Pose Estimation Using Anisotropic Spherical Gaussian
Zhiwen Cao, Dongfang Liu, Qifan Wang, Yingjie Chen
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 |
|---|---|---|---|
| use rotation matrix instead of Euler angles | direct regression, ResNet-18 · AFLW2000 (train 300W-LP) | MAE: 5.28 -> 4.31 | Improved |
| use rotation matrix instead of quaternion | direct regression, ResNet-18 · AFLW2000 (train 300W-LP) | MAE: 4.66 -> 4.31 | Improved |
| use rotation matrix instead of Lie algebra | direct regression, ResNet-18 · BIWI 30% (train BIWI 70%) | MAE: 3.26 -> 3.08 | Improved |
| add spherical Gaussian label distribution learning to direct regression (ablation) | rotation matrix model · AFLW2000 (train 300W-LP) | MAE: 4.31 -> 3.79 | Improved |
| use anisotropic instead of isotropic spherical Gaussian (ablation) | ASG distribution · AFLW2000 (train 300W-LP) | MAE: 3.79 -> 3.64 | Improved |
| use anisotropic instead of isotropic spherical Gaussian (ablation) | ASG distribution · BIWI 30% (train BIWI 70%) | MAE: 2.93 -> 2.77 | Improved |
| use adaptive parameters instead of fixed lambda=1, eta=1 (ablation) | ASG distribution · AFLW2000 (train 300W-LP) | MAE: 3.79 -> 3.64 | Improved |
| use adaptive parameters instead of fixed lambda=5, eta=5 (ablation) | ASG distribution · BIWI full (train 300W-LP) | MAE: 3.97 -> 3.61 | Improved |
| use adaptive parameters instead of fixed lambda=1, eta=5 (ablation) | ASG distribution · AFLW2000 (train 300W-LP) | MAE: 3.92 -> 3.64 | Improved |
| add regression loss L_reg to classification loss (ablation) | ASG distribution · AFLW2000 (train 300W-LP) | MAE: 3.67 -> 3.64 | Improved |
| add classification loss L_cls to regression loss (ablation) | ASG distribution · AFLW2000 (train 300W-LP) | MAE: 4.31 -> 3.64 | Improved |
| add ASG label distribution with ResNet-18 vs img2pose | ASG (ours) · BIWI full (train 300W-LP) | MAE: 3.79 -> 3.61 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- img2pose — img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation (CVPR 2021) · better on 2 of 2 shared results
- Hopenet — HOPE-Net: A Graph-Based Model for Hand-Object Pose Estimation (CVPR 2020) · better on 2 of 2 shared results
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
- Spherical Confidence Learning for Face Recognition (CVPR 2021 · similarity 0.85)
- Label Distribution Learning on Auxiliary Label Space Graphs for Facial Expression Recognition (CVPR 2020 · similarity 0.85)
- Gaussian Label Distribution Learning for Spherical Image Object Detection (CVPR 2023 · similarity 0.84)
- Unimodal-Concentrated Loss: Fully Adaptive Label Distribution Learning for Ordinal Regression (CVPR 2022 · similarity 0.84)
- Learning Shape-Independent Transformation via Spherical Representations for Category-Level Object Pose Estimation (ICLR 2025 · similarity 0.82)
- Towards Fine-Grained HBOE with Rendered Orientation Set and Laplace Smoothing (AAAI 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.
