Research map / 3D vision / Face and shape
Unsupervised Training for 3D Morphable Model Regression
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 |
|---|---|---|---|
| Train the regression network with unlabeled-image identity features and the proposed losses | 3DMM regression network · MICC cooperative, indoor, and outdoor | point-to-plane error (mm): 1.93 -> 1.50; 2.02 -> 1.50; 1.86 -> 1.48 | Helped |
| Train the regression network with unlabeled-image identity features and the proposed losses | 3DMM regression network · MoFA-Test | identity clustering recall Top-1: 0.25 -> 0.87 | Helped |
| Train the regression network with unlabeled-image identity features and the proposed losses | 3DMM regression network · LFW | identity clustering recall Top-5: 0.002 -> 0.51 | Helped |
| Add batch distribution loss to constrain batch parameter moments to the morphable-model distribution (ablation) | 3DMM regression network · Face reconstruction quality | plausibility and color fidelity | Helped |
| Add loopback loss so the network reinterprets its rendered predictions consistently (ablation) | 3DMM regression network · Face reconstruction quality | feature realism | Helped |
| Use identity loss across three randomly selected rendered views instead of one frontal view (ablation) | 3DMM regression network · Face reconstruction under expression and lighting variation | robustness | Helped |
| Add a shallow network to transform rendered-face encoder outputs before identity comparison (ablation) | 3DMM regression network · Face recognizability and output realism | output realism | Hurt |
| Use deeper decoder layers instead of the two-layer decoder (ablation) | 3DMM regression network · Face reconstruction | performance | No clear effect |
| Use the 1024-D avgpool FaceNet layer as decoder input instead of the final 128-D normalizing layer (ablation) | 3DMM regression network · 3DMM parameter prediction | decoder effectiveness | Helped |
Closest papers by meaning
- Regressing Robust and Discriminative 3D Morphable Models With a Very Deep Neural Network (CVPR 2017 · similarity 0.91)
- MVF-Net: Multi-View 3D Face Morphable Model Regression (CVPR 2019 · similarity 0.90)
- Unsupervised Depth Estimation, 3D Face Rotation and Replacement (NeurIPS 2018 · similarity 0.90)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.89)
- Nonlinear 3D Face Morphable Model (CVPR 2018 · similarity 0.89)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.87)
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.