Research map / 3D vision / NeRF and radiance fields
Benchmarking and Analyzing 3D-aware Image Synthesis with a Modularized Codebase
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 different single point embedders (MLP, volume, or tri-plane) (ablation) | EG3D · FFHQ | FID: 5.15; 4.65; 4.72 | No clear effect |
| Combine multiple point embedders rather than use a single type (ablation) | EG3D · FFHQ | FID: 4.62 | No clear effect |
| Use a bi-plane instead of a tri-plane (ablation) | EG3D · FFHQ | FID: 4.72 -> 4.38 | Helped |
| Increase MLP point-feature decoder depth from 4 to 16 (ablation) | EG3D with MLP point features · FFHQ | FID: 17.22 -> 5.15 | Helped |
| Increase tri-plane decoder depth from 2 to 8 (ablation) | EG3D with tri-plane features · FFHQ | FID: 4.72 -> 5.58 | Hurt |
| Use ReLU instead of SIREN with an upsampler (ablation) | EG3D · FFHQ 256×256 | FID: 11.66 -> 7.39 | Helped |
| Use SIREN instead of ReLU without an upsampler (ablation) | EG3D · FFHQ 64×64 | FID: 7.30 -> 6.58 | Helped |
| Replace density with SDF (ablation) | EG3D with MLP, volume, or tri-plane features · FFHQ | FID: 5.15 -> 8.87; 4.65 -> 7.27; 4.72 -> 13.31 | Hurt |
| Remove the upsampler (ablation) | EG3D with tri-plane features · FFHQ 256×256 | FID: 4.72 -> 6.86 | Hurt |
| Remove the upsampler (ablation) | EG3D with tri-plane features · FFHQ 256×256 | RE: 0.111 -> 0.104 | Helped |
| Use ground-truth poses instead of random pose distribution (ablation) | EG3D with MLP, volume, or tri-plane features · FFHQ | FID: 14.56 -> 5.15; 10.47 -> 4.65; 15.18 -> 4.72 | Helped |
| Replace random poses with accurate pose distribution (ablation) | EG3D with MLP, volume, or tri-plane features · FFHQ | FID: 14.56 -> 9.96; 10.47 -> 7.34; 15.18 -> 5.45 | Helped |
Closest papers by meaning
- Improving 3D-aware Image Synthesis with A Geometry-aware Discriminator (NeurIPS 2022 · similarity 0.92)
- 3D-Aware Image Synthesis via Learning Structural and Textural Representations (CVPR 2022 · similarity 0.91)
- GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis (NeurIPS 2020 · similarity 0.91)
- NerfBaselines: Consistent and Reproducible Evaluation of Novel View Synthesis Methods (NeurIPS 2025 · similarity 0.88)
- G-NeRF: Geometry-enhanced Novel View Synthesis from Single-View Images (CVPR 2024 · similarity 0.88)
- CAD: Photorealistic 3D Generation via Adversarial Distillation (CVPR 2024 · 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.