Research map / 3D vision / Human pose estimation
Heatmap Regression without Soft-Argmax for Facial Landmark Detection
Chiao-An Yang, Raymond A. Yeh
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 |
|---|---|---|---|
| replace Soft-argmax l2 loss (STAR) with structured prediction loss, no label smoothing (ablation) | heatmap regression · WFLW full | FR: 2.32 -> 2.23 | No clear effect |
| add image-aware label smoothing to structured loss (ablation) | structured loss · WFLW full | FR: 2.23 -> 1.84 | Improved |
| add image-aware label smoothing to structured loss (ablation) | structured loss · WFLW full | NME: 4.02 -> 3.99 | Improved |
| make label smoothing image-aware instead of image-unaware (ablation) | structured loss · WFLW full | FR: 2.12 -> 1.84 | Improved |
| add image-aware label smoothing to STAR loss (ablation) | STAR · WFLW full | FR: 2.32 -> 2.12 | Improved |
| use smooth-l1 margin instead of no margin (ablation) | structured loss · WFLW full | FR: 2.16 -> 1.84 | Improved |
| use smooth-l1 margin instead of l2 (ablation) | structured loss · WFLW full | FR: 2.00 -> 1.84 | Improved |
| use smooth-l1 margin instead of l1 (ablation) | structured loss · WFLW full | FR: 2.04 -> 1.84 | Improved |
| add image-aware label smoothing (ablation) | structured loss · WFLW-L (large pose) | NME: 6.68 -> 6.58 | Improved |
| add image-aware label smoothing (ablation) | structured loss · WFLW-M (make-up) | NME: 3.93 -> 3.89 | Improved |
| add image-aware label smoothing (ablation) | structured loss · WFLW-B (blur) | NME: 4.56 -> 4.57 | No clear effect |
| replace Soft-argmax with structured prediction loss plus label smoothing | STAR · WFLW full | NME: 4.02 -> 3.99 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- STAR — STAR: Sparse Trained Articulated Human Body Regressor (ECCV 2020) · better on 11 of 14 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
- Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression (ICCV 2019 · similarity 0.88)
- Learning Robust Facial Landmark Detection via Hierarchical Structured Ensemble (ICCV 2019 · similarity 0.88)
- Pixel-in-Pixel Net: Towards Efficient Facial Landmark Detection in the Wild (IJCV 2021 · similarity 0.88)
- Heatmap Regression via Randomized Rounding (TPAMI 2021 · similarity 0.87)
- Brute-Force Facial Landmark Analysis With a 140,000-Way Classifier (AAAI 2018 · similarity 0.87)
- T-FAKE: Synthesizing Thermal Images for Facial Landmarking (CVPR 2025 · similarity 0.84)
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.
