Research map / 3D vision / NeRF and radiance fields
GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis
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 2D image generation with GRAF's conditional radiance field | GRAF · Cars, 64² | FID: 66 -> 30 | Helped |
| replace voxel-based HoloGAN with GRAF's continuous radiance field | GRAF · Cars, 128² | FID: 180 -> 41 | Helped |
| sample at higher resolution during inference rather than bilinearly upsample | GRAF trained at 128² · Cars, 256² | FID: 91 -> 74 | Helped |
| remove 3D convolutional layers from HoloGAN (ablation) | HoloGAN · Cars, 128² | FID: 211 -> 180 | Helped |
| use multi-scale patch discrimination instead of full-image discrimination (ablation) | GRAF · CelebA, 64², dim/2 | FID: 77 -> 28 | Helped |
| use multi-scale patch discrimination instead of full-image discrimination (ablation) | GRAF · Cars, 64², dim/2 | FID: 29 -> 32 | Hurt |
| restrict patch sampling to local scale s=1 (ablation) | GRAF · Cars, 128² | FID: 41 -> 277 | Hurt |
| restrict patch sampling to local scale s=1 (ablation) | GRAF · CelebA, 128² | FID: 35 -> 153 | Hurt |
| vary focal length within 0.7 f_data to 1.0 f_data (ablation) | GRAF · Cars, fixed radius 10 | FID: 40; 36; 41 | No clear effect |
| use a very small focal length of 0.3 f_data (ablation) | GRAF · Cars, fixed radius 10 | FID: 84 | Hurt |
| use orthographic projection (ablation) | GRAF · Cars, fixed radius 10 | FID: 53 | No clear effect |
| remove positional encoding from the radiance-field input (ablation) | GRAF · supplementary ablation | fitting high-frequency signals | Hurt |
Closest papers by meaning
- Towards Unsupervised Learning of Generative Models for 3D Controllable Image Synthesis (CVPR 2020 · similarity 0.91)
- GRAM: Generative Radiance Manifolds for 3D-Aware Image Generation (CVPR 2022 · similarity 0.91)
- DisCoScene: Spatially Disentangled Generative Radiance Fields for Controllable 3D-Aware Scene Synthesis (CVPR 2023 · similarity 0.91)
- SinGRAF: Learning a 3D Generative Radiance Field for a Single Scene (CVPR 2023 · similarity 0.91)
- Pix2NeRF: Unsupervised Conditional p-GAN for Single Image to Neural Radiance Fields Translation (CVPR 2022 · similarity 0.91)
- Latent Radiance Fields with 3D-aware 2D Representations (ICLR 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.