Research map / Image generation / Editing and style transfer
NeRFInvertor: High Fidelity NeRF-GAN Inversion for Single-Shot Real Image Animation
Yu Yin, Kamran Ghasedi, HsiangTao Wu, Jiaolong Yang, Xin Tong, Yun Fu
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 |
|---|---|---|---|
| Fine-tune the generator with image losses and implicit and explicit neighborhood regularization | NeRFInvertor with AniFaceGAN · FFHQ test | Novel-view FID: 50.20 -> 45.89 | Improved |
| Fine-tune the generator with image losses and implicit and explicit neighborhood regularization | NeRFInvertor with AniFaceGAN · CelebA-HQ test | Novel-view ID: 0.73 -> 0.77 | Improved |
| Fine-tune the generator with image losses and implicit and explicit neighborhood regularization | NeRFInvertor with GRAM · FFHQ test | Novel-view FID: 45.94 -> 43.58 | Improved |
| Fine-tune the generator with image losses and implicit and explicit neighborhood regularization | NeRFInvertor with GRAM · CelebA-HQ test | Novel-view FID: 30.95 -> 31.11 | Hurt |
| Add implicit geometrical regularization to image-space supervision (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view FID: 38.61 -> 37.83 | Improved |
| Add implicit geometrical regularization to image-space supervision (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view ID: 0.66 -> 0.73 | Improved |
| Add explicit geometrical regularization after implicit regularization (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view FID: 37.83 -> 34.68 | Improved |
| Add explicit geometrical regularization after implicit regularization (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view ID: 0.73 -> 0.76 | Improved |
| Add masked regularizations to the explicit-regularization variant (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view FID: 34.68 -> 34.07 | Improved |
| Add masked regularizations to the explicit-regularization variant (ablation) | NeRFInvertor ablation · CelebA-HQ test | Novel-view ID: 0.76 -> 0.77 | Improved |
| Set neighborhood distance too small (α<=1) (ablation) | NeRFInvertor · Novel-view animation | identity preservation | Hurt |
| Set neighborhood distance too large (α>=10) (ablation) | NeRFInvertor · Novel-view animation | 3D geometry | Hurt |
Closest papers by meaning
- FaceDNeRF: Semantics-Driven Face Reconstruction, Prompt Editing and Relighting with Diffusion Models (NeurIPS 2023 · similarity 0.88)
- GAN Inversion: A Survey (TPAMI 2022 · similarity 0.88)
- AniFaceGAN: Animatable 3D-Aware Face Image Generation for Video Avatars (NeurIPS 2022 · similarity 0.88)
- G-NeRF: Geometry-enhanced Novel View Synthesis from Single-View Images (CVPR 2024 · similarity 0.87)
- Diffusion-driven GAN Inversion for Multi-Modal Face Image Generation (CVPR 2024 · similarity 0.86)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (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.
