Research map / 3D vision / Face and shape
RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild
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 |
|---|---|---|---|
| jointly predict five facial landmarks and 3D mesh vertices (ablation) | RetinaFace detector · WIDER FACE Hard test subset | average AP: 54.20 -> 55.02 | Helped |
| add deformable context modules (ablation) | RetinaFace detector · WIDER FACE Hard test subset | average AP: 52.65 -> 53.36 | Helped |
| add cascade regression (ablation) | RetinaFace detector with DCN · WIDER FACE Hard test subset | average AP: 53.36 -> 54.20 | Helped |
| add five facial landmark regression (ablation) | RetinaFace detector with cascade regression · WIDER FACE Hard test subset | average AP: 54.20 -> 54.73 | Helped |
| add 1k 3D vertex regression (ablation) | RetinaFace detector with cascade regression · WIDER FACE Hard test subset | average AP: 54.20 -> 54.55 | Helped |
| add topology constraints to 1k 3D vertex regression (ablation) | RetinaFace detector · WIDER FACE Hard test subset | average AP: 54.55 -> 54.62 | Helped |
| add a 3D mesh regression branch (ablation) | RetinaFace-R50 · WIDER FACE validation set | AUC: 55.66% -> 58.54% | Helped |
| add a 3D mesh regression branch (ablation) | RetinaFace-R50 · WIDER FACE validation set | Failure Rate (threshold at 10%): 10.25% -> 9.82% | Helped |
| add mesh topology constraints to direct vertex regression (ablation) | RetinaFace on AFLW2000-3D · AFLW2000-3D | NME [0°,30°]: 2.77 -> 2.72 | Helped |
| jointly train five-point landmark regression and mesh regression (ablation) | RetinaFace on AFLW2000-3D · AFLW2000-3D | NME [0°,30°]: 2.72 -> 2.57 | Helped |
| add mesh topology constraints to direct vertex regression (ablation) | RetinaFace yaw estimation · AFLW2000-3D [0°,30°] | yaw angle error: 3.79 ± 3.08 -> 3.69 ± 2.99 | Helped |
| use direct image-plane vertex regression instead of indirect UV-coordinate or 3DMM-parameter regression | RetinaFace · AFLW2000-3D | NME (68 points, mean): PRN 3.62; MS-SFN 3.89; L5pts+mesh 3.48 | Helped |
Closest papers by meaning
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.88)
- 3FabRec: Fast Few-Shot Face Alignment by Reconstruction (CVPR 2020 · similarity 0.88)
- Learning to Detect 3D Facial Landmarks via Heatmap Regression with Graph Convolutional Network (AAAI 2022 · similarity 0.87)
- Look at Boundary: A Boundary-Aware Face Alignment Algorithm (CVPR 2018 · similarity 0.87)
- Joint Face Detection and Facial Motion Retargeting for Multiple Faces (CVPR 2019 · similarity 0.87)
- 3D Face Tracking from 2D Video through Iterative Dense UV to Image Flow (CVPR 2024 · similarity 0.86)
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.