Research map / 3D vision / Face and shape
Learning Flexibly Distributional Representation for Low-quality 3D Face Recognition
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 |
|---|---|---|---|
| Replace deterministic face features with a flexible posterior learned using CNF | Led3D backbone · Lock3DFace Protocol I | accuracy (AVG): 54.28 -> 58.68 | Helped |
| Use a flexible CNF posterior instead of the Gaussian posterior (ablation) | Ours · Lock3DFace Protocol II, divisions A-D | accuracy (AVG), video and frame evaluation: A video: 55.48 -> 58.72; A frame: 53.84 -> 55.52; B video: 54.12 -> 57.65; B frame: 52.79 -> 55.13; C video: 67.61 -> 70 | Helped |
| Use a flexible CNF posterior instead of the Gaussian posterior (ablation) | Ours · Lock3DFace Protocol II, pose subset in division B | accuracy: 28.69 -> 36.30 | Hurt |
| Use a flexible CNF posterior instead of the Gaussian posterior (ablation) | Ours · Lock3DFace Protocol II, time subset in division D | accuracy: 55.00 -> 60.00 | Helped |
| Use a flexible CNF posterior instead of the Gaussian posterior (ablation) | Ours · Lock3DFace Protocol II | accuracy: B frame time subset: 23.40 -> 23.13; D video time subset: 55.00 -> 60.00 | Mixed |
| Use a flexible CNF posterior instead of the Gaussian posterior (ablation) | Ours · Lock3DFace Protocol II | accuracy: A video AVG: 55.48 -> 58.72; D frame AVG: 70.26 -> 71.89 | Helped |
| Use flexible distributional representation with CNF rather than Led3D | 3D face recognition model · IIIT-D | accuracy: 74.27 -> 80.28 | Helped |
| Use flexible distributional representation with CNF rather than Led3D | 3D face recognition model · BU3D-FE and BU4D-FE with synthetic Gaussian noise | accuracy: BU3D-FE: 95.08 vs 96.88 (variance 4), 74.66 vs 82.33 (variance 32); BU4D-FE: 90.07 vs 94.87 (variance 4), 50.51 vs 61.32 | Helped |
| Use flexible distributional representation with CNF rather than Led3D | 3D face recognition model · BU3D-FE and BU4D-FE cross-quality recognition | accuracy | Helped |
| Increase synthetic noise intensity | Ours versus Led3D · BU3D-FE and BU4D-FE synthetic noisy data | accuracy: BU3D-FE variance 64: 55.37 -> 60.37; BU4D-FE variance 64: 28.73 -> 36.80 | Mixed |
| Use MLS from the learned posterior as the face feature | Ours · BU3D-FE high-low quality feature pairs | feature distance | Helped |
| Use flexible posterior rather than Gaussian posterior (ablation) | Ours · Feature clustering visualization | feature compactness | Helped |
Closest papers by meaning
- Led3D: A Lightweight and Efficient Deep Approach to Recognizing Low-Quality 3D Faces (CVPR 2019 · similarity 0.86)
- A Robust Multilinear Model Learning Framework for 3D Faces (CVPR 2016 · similarity 0.84)
- Regressing Robust and Discriminative 3D Morphable Models With a Very Deep Neural Network (CVPR 2017 · similarity 0.84)
- Gaussian Differentially Private Human Faces Under a Face Radial Curve Representation (ICLR 2025 · similarity 0.83)
- RAFaRe: Learning Robust and Accurate Non-parametric 3D Face Reconstruction from Pseudo 2D&3D Pairs (AAAI 2023 · similarity 0.83)
- Blind Face Restoration under Extreme Conditions: Leveraging 3D-2D Prior Fusion for Superior Structural and Texture Recovery (AAAI 2024 · similarity 0.83)
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.