Research map / Detection and segmentation / CNN features
Towards Accurate Facial Landmark Detection via Cascaded Transformers
Check 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 |
|---|---|---|---|
| Use the parallel decoder to update image features alongside landmark coordinates (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.05 | Helped |
| Initialize queries from image features rather than random learned embeddings (ablation) | DTLD · WFLW-Full | NME (%): 4.19 -> 4.08 | Helped |
| Add self-attention among landmark queries (ablation) | DTLD · WFLW-Full | NME (%): 4.17 -> 4.08 | Helped |
| Use a deeper ResNet-50 or ResNet-101 backbone instead of ResNet-18 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.07 | No clear effect |
| Increase feature dimension from 256 to 512 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.07 | No clear effect |
| Reduce feature dimension from 256 to 64 or 128 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.47; 4.08 -> 4.29 | Hurt |
| Vary sampling points from 4 to 2 or 6 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.11; 4.08 -> 4.09 | No clear effect |
| Use 4 attention heads instead of 8 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.16 | Hurt |
| Use 16 attention heads instead of 8 (ablation) | DTLD · WFLW-Full | NME (%): 4.08 -> 4.07 | No clear effect |
| Add encoder layers to DTLD (ablation) | DTLD · WFLW-Full | NME (%): 4.076 -> 4.066 -> 4.047 -> 4.028 -> 3.999 | Helped |
| Increase decoder layers from 1 to 6 (ablation) | DTLD · WFLW-Full | NME (%): 4.417 -> 4.064 | Helped |
| Use parallel decoder instead of serial encoder layers at similar parameter count (ablation) | DTLD · WFLW-Full | NME (%): 4.028 -> 4.054 | Helped |
Closest papers by meaning
- Facial Landmarks Detection by Self-Iterative Regression Based Landmarks-Attention Network (AAAI 2018 · similarity 0.90)
- Dynamic Attention-Controlled Cascaded Shape Regression Exploiting Training Data Augmentation and Fuzzy-Set Sample Weighting (CVPR 2017 · similarity 0.88)
- Self-Reinforced Cascaded Regression for Face Alignment (AAAI 2018 · similarity 0.87)
- Continuous Landmark Detection With 3D Queries (CVPR 2023 · similarity 0.87)
- POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search (AAAI 2025 · similarity 0.85)
- Generalizable Face Landmarking Guided by Conditional Face Warping (CVPR 2024 · 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.