Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Face Alignment With Kernel Density Deep Neural Network
Lisha Chen, Hui Su, Qiang Ji
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 Gaussian kernel density loss instead of softlabel loss | Hourglass cascade · 300W-test | NME / NLL: NME 2.32 -> 2.21; NLL 4.67 -> 2.93 | Improved |
| use Gaussian kernel density loss instead of softlabel loss | Hourglass cascade · COFW-68 test | NME / NLL: NME 2.92 -> 2.73; NLL 5.27 -> 3.21 | Improved |
| use Gaussian kernel density loss instead of softlabel loss | Hourglass cascade · Menpo-frontal | NME / AUC: NME 2.27 -> 2.01; AUC 67.4 -> 71.1 | Improved |
| use Gaussian kernel instead of uniform kernel (ablation) | KDN · 300W-test | NME: 2.38 -> 2.21 | Improved |
| use Gaussian kernel instead of uniform kernel (ablation) | KDN · COFW-68 test | NME: 2.92 -> 2.73 | Improved |
| use KDN loss instead of softlabel in a single stage (ablation) | single-stage Hourglass · 300W-test | NME / NLL: NME 2.58 -> 2.52; NLL 4.79 -> 3.01 | Improved |
| use Gaussian KDN instead of softlabel | Hourglass cascade · COFW-29 | NME / NLL: NME 2.51 -> 2.28; NLL 6.13 -> 3.19 | Improved |
| use Gaussian KDN instead of softlabel | Hourglass cascade · AFLW-full | NME / NLL: NME 2.87 -> 2.80; NLL 5.75 -> 3.56 | Improved |
| use KDN uncertainty instead of softlabel pseudo-variance for occlusion detection | KDN-Gaussian · COFW-68 | occlusion precision/recall: precision 56/40 -> 70/40 | Improved |
| add multi-stage cascade with probability propagation (ablation) | KDN · 300W-test | NME and uncertainty | Improved |
| use KDN-Gaussian instead of Gaussian for AU intensity | facial AU intensity · BP4D | MAE: 0.748 -> 0.757 | Hurt |
| use KDN-Gaussian instead of KDN-Uniform for AU intensity (ablation) | facial AU intensity · BP4D | ICC: 0.559 -> 0.588 | Improved |
Closest papers by meaning
- How Far Are We From Solving the 2D & 3D Face Alignment Problem? (And a Dataset of 230,000 3D Facial Landmarks) (ICCV 2017 · similarity 0.86)
- Direct Shape Regression Networks for End-to-End Face Alignment (CVPR 2018 · similarity 0.86)
- Robust Face Alignment via Deep Progressive Reinitialization and Adaptive Error-driven Learning (TPAMI 2021 · similarity 0.86)
- Large-Pose Face Alignment via CNN-Based Dense 3D Model Fitting (CVPR 2016 · similarity 0.85)
- Face Alignment Across Large Poses: A 3D Solution (CVPR 2016 · similarity 0.85)
- Pose-Invariant Face Alignment With a Single CNN (ICCV 2017 · 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.
