Research map / Graph learning / Knowledge and temporal graphs
Structured Landmark Detection via Topology-Adapting Deep Graph Learning
Weijian Li, Yuhang Lu, Kang Zheng, Haofu Liao, Chihung Lin, Jiebo Luo, Chi-Tung Cheng, Jing Xiao, Le Lu, Chang-Fu Kuo, Shun Miao
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 |
|---|---|---|---|
| add learned graph connectivity instead of uniform (with shape feature) (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.12 -> 3.04 | Improved |
| add information exchange (uniform connectivity) instead of self-connectivity only (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.31 -> 3.16 | Improved |
| add shape feature to graph signal with self connectivity (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.31 -> 3.16 | Improved |
| add shape feature to graph signal with learned connectivity (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.08 -> 3.04 | Improved |
| use perspective instead of affine transformation in GCN-global (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.13 -> 3.04 | Improved |
| use GCN-global instead of a CNN block (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.12 -> 3.04 | Improved |
| use GCN-local instead of an MLP block (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.18 -> 3.04 | Improved |
| increase GCN-local steps from 1 to 3 (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.24 -> 3.04 | Improved |
| increase GCN-local steps from 3 to 7 (ablation) | DAG HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.04 -> 3.11 | Hurt |
| use DAG graph regression instead of local feature regression (ablation) | HRNet18 · 300W Full set (inter-ocular) | NME (%): 3.72 -> 3.04 | Improved |
| use DAG instead of global feature regression with VGG16 backbone (ablation) | VGG16 · 300W Full set (inter-ocular) | NME (%): 4.66 -> 3.66 | Improved |
| add WING loss to DAG | DAG · 300W Full set (inter-pupil) | NME (%): 4.27 -> 4.21 | Improved |
Closest papers by meaning
- Facial Landmark Detection with Tweaked Convolutional Neural Networks (TPAMI 2017 · similarity 0.87)
- Learning to Detect 3D Facial Landmarks via Heatmap Regression with Graph Convolutional Network (AAAI 2022 · similarity 0.87)
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · similarity 0.86)
- Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training (AAAI 2025 · similarity 0.86)
- Deep Structured Prediction for Facial Landmark Detection (NeurIPS 2019 · similarity 0.85)
- Spot the Difference: Detection of Topological Changes via Geometric Alignment (NeurIPS 2021 · 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.
