Research map / 3D vision / Face and shape
$$E^{3}DGE$$: Self-Supervised Geometry-Aware Encoder for Style-Based 3D GAN Inversion
Yushi Lan, Xuyi Meng, Shuai Yang, Chen Change Loy, Bo Dai
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 local pixel-aligned features to synthetic-trained global encoder (ablation) | E3DGE · CelebA-HQ source view | identity similarity: .369 -> .953 | Improved |
| add local pixel-aligned features without novel-view training (ablation) | E3DGE · novel view reconstruction | identity similarity: .770 -> .608 | Hurt |
| add 3D alignment to local features (ablation) | E3DGE · novel view reconstruction | MAE: .282 -> .133 | Improved |
| add 2D alignment to local features (ablation) | E3DGE · novel view reconstruction | identity similarity: .608 -> .904 | Improved |
| use hybrid alignment instead of 3D alignment alone | E3DGE · source view reconstruction | identity similarity: .818 -> .883 | Hurt |
| use hybrid alignment instead of 2D alignment alone (ablation) | E3DGE · novel view reconstruction | MAE: .178 -> .131 | Improved |
| add geometry loss LO geo (ablation) | e4e StyleSDF · NoW validation set | mean error: 3.40 -> 2.11 | Improved |
| add geometry loss LF geo (ablation) | e4e StyleSDF with LO geo · NoW validation set | mean error: 2.11 -> 2.09 | Improved |
| add code loss Lcode (ablation) | e4e StyleSDF with geometry losses · NoW validation set | mean error: 2.09 -> 2.06 | Improved |
| train on fully synthetic data only versus pSp (ablation) | pSp-like encoder · CelebA-HQ source view | MAE: .150 -> .245 | Hurt |
| use E3DGE instead of e4e on StyleSDF | E3DGE · novel view reconstruction | identity similarity: .271 -> .903 | Improved |
| use E3DGE encoder instead of PTI optimization | E3DGE · NoW validation set | mean error: 3.54 -> 2.06 | Improved |
Closest papers by meaning
- Self-Supervised Geometry-Aware Encoder for Style-Based 3D GAN Inversion (CVPR 2023 · similarity 0.99)
- 3D GAN Inversion With Facial Symmetry Prior (CVPR 2023 · similarity 0.93)
- Make Encoder Great Again in 3D GAN Inversion through Geometry and Occlusion-Aware Encoding (ICCV 2023 · similarity 0.92)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (CVPR 2024 · similarity 0.91)
- Dual Encoder GAN Inversion for High-Fidelity 3D Head Reconstruction from Single Images (NeurIPS 2024 · similarity 0.91)
- WarpGAN: Warping-Guided 3D GAN Inversion with Style-Based Novel View Inpainting (NeurIPS 2025 · similarity 0.89)
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.
