Research map / Detection and segmentation / CNN features
STAR Loss: Reducing Semantic Ambiguity in Facial Landmark Detection
Read the paperCheck 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 STAR loss to heatmap regression | Hourglass facial landmark detector · WFLW-Full | NME: 4.08 -> 4.02 | Helped |
| Add STAR loss to heatmap regression | Hourglass facial landmark detector · 300W Challenging | NME: 4.50 -> 4.32 | Helped |
| Add STAR loss to L1, L2, smooth-L1, or Wing regression (ablation) | Basic Hourglass · COFW, 300W, and WFLW | NME: Average improvements of 0.34, 0.19, and 0.25 on COFW, 300W, and WFLW | Helped |
| Add STAR loss to L1 regression (ablation) | Hourglass trained on 300W · 300VW Category-C | NME: 8.71/8.55 | Helped |
| Add STAR loss to L2 regression (ablation) | Hourglass trained on 300W · 300VW Category-C | NME: 8.44/8.27 | Helped |
| Add STAR loss to smooth-L1 regression (ablation) | Hourglass trained on 300W · 300VW Category-C | NME: 8.69/8.42 | Helped |
| Use eigenvalue value restriction rather than no restriction (ablation) | STAR loss · WFLW | NME: 4.60 -> 4.17 | Helped |
| Use detach restriction rather than no restriction (ablation) | STAR loss · WFLW | NME: 4.60 -> 4.20 | Helped |
| Add AAM to STAR with value restriction (ablation) | STAR loss · WFLW | NME: 4.17 -> 4.02 | Helped |
| Vary value-restriction weight from w=1 to w=2, 3, 5, 10, or 20 (ablation) | STAR with smooth-L1 · WFLW | NME: 4.17; 4.20; 4.18; 4.18; 4.18; 4.20 | No clear effect |
| Use STAR loss instead of regression baseline | Hourglass facial landmark detector · WFLW test set | prediction variance: 0.79 -> 0.69 | Helped |
| Replace ADL with STAR loss (ablation) | ADNet · WFLW Full | NME: 4.14 -> 4.02 | Helped |
Closest papers by meaning
- Flaws can be Applause: Unleashing Potential of Segmenting Ambiguous Objects in SAM (NeurIPS 2024 · similarity 0.83)
- Semantic-Aware Transformation-Invariant RoI Align (AAAI 2024 · similarity 0.83)
- SC-Tune: Unleashing Self-Consistent Referential Comprehension in Large Vision Language Models (CVPR 2024 · similarity 0.82)
- AdapToPASS: Ambiguity-aware Adaptive Spherical Transformer for Panoramic Semantic Segmentation (NeurIPS 2026 · similarity 0.82)
- A Black-Box Evaluation Framework for Semantic Robustness in Bird’s Eye View Detection (AAAI 2025 · similarity 0.82)
- POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search (AAAI 2025 · 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.