Research map / 3D vision / Face and shape
Continuous Landmark Detection With 3D Queries
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 |
|---|---|---|---|
| Train ConvNext+MLP with all data instead of Fake-it only (ablation) | Continuous Landmark Detection · 300-W Challenging subset | NME: 5.96 -> 4.46 | Helped |
| Train MobileNet+MLP with all data instead of Fake-it only (ablation) | Continuous Landmark Detection · 300-W Common subset | NME: 3.62 -> 3.20 | Helped |
| Train MobileNet+MLP with all data instead of Fake-it only (ablation) | Continuous Landmark Detection · 300-W Challenging subset | NME: 7.85 -> 6.14 | Helped |
| Train ConvNext+Transformer with all data instead of Fake-it only (ablation) | Continuous Landmark Detection · 300-W Common subset | NME: 2.99 -> 2.87 | Helped |
| Train ConvNext+Transformer with all data instead of Fake-it only (ablation) | Continuous Landmark Detection · 300-W Challenging subset | NME: 4.71 -> 4.42 | Helped |
| Use a Transformer query predictor instead of MLP with ConvNext and all data (ablation) | Continuous Landmark Detection · Dense 2D landmarks | Dense NME: 1.4 -> 1.07 | Helped |
| Use a Transformer query predictor instead of MLP with ConvNext and all data (ablation) | Continuous Landmark Detection · 300-W Challenging subset | NME: 4.46 -> 4.42 | Helped |
| Use a Transformer query predictor instead of MLP with ConvNext and all data (ablation) | Continuous Landmark Detection · 300-W Common subset | NME: 3.16 -> 2.87 | Helped |
| Use MobileNet instead of ConvNext with MLP and all data (ablation) | Continuous Landmark Detection · Dense 2D landmarks | Dense NME: 1.4 -> 1.95 | Hurt |
| Increase landmark count from 68 to 50K (ablation) | 3D face reconstruction · 3D face reconstruction | Mean error (mm): 2.88 -> 2.34 | Helped |
| Use 3 views instead of the single-view 68-landmark setting (ablation) | 3D face reconstruction · 3D face reconstruction | Mean error (mm): 2.88 -> 2.76 | Helped |
| Add color augmentation during training (ablation) | ConvNext+MLP · 300-W Common test set | NME: 4.27 -> 4.24 | Helped |
Closest papers by meaning
- Learning to Detect 3D Facial Landmarks via Heatmap Regression with Graph Convolutional Network (AAAI 2022 · similarity 0.90)
- 3D-Aware Facial Landmark Detection via Multi-View Consistent Training on Synthetic Data (CVPR 2023 · similarity 0.90)
- Deep Structured Prediction for Facial Landmark Detection (NeurIPS 2019 · similarity 0.89)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.88)
- Facial Landmarks Detection by Self-Iterative Regression Based Landmarks-Attention Network (AAAI 2018 · similarity 0.88)
- Unsupervised Discovery of Facial Landmarks and Head Pose (CVPR 2025 · similarity 0.88)
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.