Research map / Point clouds / Point cloud classification
Learning to Detect 3D Facial Landmarks via Heatmap Regression with Graph Convolutional Network
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 |
|---|---|---|---|
| Use the proposed PAConv-based graph network and heatmap pipeline instead of existing methods | 3D facial landmark detection · BU-3DFE | ME: 2.32 -> 1.97 | Helped |
| Use the proposed PAConv-based graph network and heatmap pipeline instead of existing methods | 3D facial landmark detection · FRGCv2 | ME: 2.84 -> 2.54 | Helped |
| Use local surface unfolding and registration instead of soft-argmax (ablation) | 3D heatmap landmark prediction · BU-3DFE | ME: 2.61 -> 1.97 | Helped |
| Use local surface unfolding and registration instead of soft-argmax (ablation) | 3D heatmap landmark prediction · FRGCv2 | ME: 3.42 -> 2.54 | Helped |
| Use local surface unfolding and registration instead of argmax (ablation) | 3D heatmap landmark prediction · BU-3DFE | ME: 2.26 -> 1.97 | Helped |
| Use local surface unfolding and registration instead of argmax (ablation) | 3D heatmap landmark prediction · FRGCv2 | ME: 2.84 -> 2.54 | Helped |
| Add heatmap weighting to local surface unfolding (ablation) | 3D heatmap landmark prediction · BU-3DFE | ME: 1.99 -> 1.97 | Helped |
| Add heatmap weighting to local surface unfolding (ablation) | 3D heatmap landmark prediction · FRGCv2 | ME: 2.59 -> 2.54 | Helped |
| Use the customized PAConv-based network rather than standard point-cloud backbones (ablation) | 3D landmark heatmap extraction · BU-3DFE | ME: 2.08 -> 1.97 | Helped |
| Set graph neighbors k to 30 rather than 20, 25, or 35 (ablation) | Proposed 3D landmark detector · BU-3DFE and FRGCv2 | ME: BU-3DFE: 3.54, 2.13, 1.97, 2.03; FRGCv2: 3.62, 2.76, 2.54, 2.64 | Mixed |
| Set local unfolding point count r to 10 rather than 15, 12, or 8 (ablation) | Proposed 3D landmark detector · BU-3DFE and FRGCv2 | ME: BU-3DFE: 1.98, 2.13, 1.97, 2.00; FRGCv2: 2.59, 2.60, 2.54, 2.61 | Mixed |
Closest papers by meaning
- Continuous Landmark Detection With 3D Queries (CVPR 2023 · similarity 0.90)
- Disentangling 3D Pose in a Dendritic CNN for Unconstrained 2D Face Alignment (CVPR 2018 · similarity 0.90)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.89)
- Deep Structured Prediction for Facial Landmark Detection (NeurIPS 2019 · similarity 0.89)
- Unsupervised Discovery of Facial Landmarks and Head Pose (CVPR 2025 · similarity 0.88)
- T-FAKE: Synthesizing Thermal Images for Facial Landmarking (CVPR 2025 · 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.