Research map / 3D vision / 3D generation
3D generation on ImageNet
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 |
|---|---|---|---|
| Use 3DGP rather than EG3D or EpiGRAF | 3D synthesis on SDIP Dogs40k, SDIP Elephants40k, and LSUN Horses40k · SDIP Dogs40k; SDIP Elephants40k; LSUN Horses40k | NFS: 34.35; 32.8; 30.4 | Helped |
| Use 3DGP rather than EG3D or EpiGRAF | 3D synthesis on SDIP Dogs40k and SDIP Elephants40k · SDIP Dogs40k; SDIP Elephants40k | FID2k: 8.74; 5.79 | Helped |
| Use 3DGP rather than EG3D or EpiGRAF | 3D synthesis on LSUN Horses40k · LSUN Horses40k | FID2k: 4.86 | Mixed |
| Train 3DGP instead of the strongest 2D baseline | ImageNet 256² synthesis · ImageNet 256² | FID: 2.30 -> 19.71 | Hurt |
| Add adversarial depth supervision and a camera generator to bare 3DGP (ablation) | 3DGP · SDIP Dogs40k | NFS: 1.42 -> 36.9 | Helped |
| Use ADS with P(non-adapted depth)=0.0 (ablation) | 3DGP · SDIP Dogs40k | NFS: 1.42 -> 3.14 | Helped |
| Use ADS with P(non-adapted depth)=0.25 (ablation) | 3DGP · SDIP Dogs40k | NFS: 1.42 -> 33.21 | Helped |
| Use ADS with P(non-adapted depth)=0.5 (ablation) | 3DGP · SDIP Dogs40k | NFS: 1.42 -> 36.9 | Helped |
| Use ADS with P(non-adapted depth)=1.0 (ablation) | 3DGP · SDIP Dogs40k | FID2k: 8.59 -> 12.2 | Hurt |
| Use ADS with P(non-adapted depth)=1.0 (ablation) | 3DGP · SDIP Dogs40k | NFS: 1.42 -> 27.2 | Helped |
| Distill ResNet features into the discriminator instead of using projected D (ablation) | 3DGP · ImageNet 128² @ 10M | FID: 105.9 -> 27.8 | Helped |
| Use projected D instead of the standard discriminator (ablation) | 3DGP · ImageNet 128² @ 10M | FID: 53.6 -> 105.9 | Hurt |
Closest papers by meaning
- EpiGRAF: Rethinking training of 3D GANs (NeurIPS 2022 · similarity 0.92)
- Improving 3D-aware Image Synthesis with A Geometry-aware Discriminator (NeurIPS 2022 · similarity 0.90)
- Efficient Geometry-Aware 3D Generative Adversarial Networks (CVPR 2022 · similarity 0.90)
- What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs (CVPR 2024 · similarity 0.89)
- G3DR: Generative 3D Reconstruction in ImageNet (CVPR 2024 · similarity 0.89)
- Do 2D GANs Know 3D Shape? Unsupervised 3D Shape Reconstruction from 2D Image GANs (ICLR 2021 · 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.