Research map / 3D vision / 3D generation
GAUDI: A Neural Architect for Immersive 3D Scene Generation
Miguel Ángel Bautista, Pengsheng Guo, Samira Abnar, Walter Talbott, Alexander T Toshev, Zhuoyuan Chen, Laurent Dinh, Shuangfei Zhai, Hanlin Goh, Daniel Ulbricht, Afshin Dehghan, Joshua M. Susskind
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 |
|---|---|---|---|
| break correspondence between pose and scene latents (random pose) (ablation) | GAUDI prior · VLN-CE | FID: 18.52 -> 83.66 | Hurt |
| widen denoiser base channels from 64 to 224 (ablation) | GAUDI DDPM prior · VLN-CE | FID: 104.27 -> 18.52 | Improved |
| widen denoiser base channels from 192 to 224 (ablation) | GAUDI DDPM prior · VLN-CE | FID: 18.61 -> 18.52 | No clear effect |
| prior noise scale beta from 0.1 to 0.0 or 0.2 (ablation) | GAUDI prior · VLN-CE | FID: 18.52 vs 18.48 | No clear effect |
| increase prior noise scale beta to 0.5 (ablation) | GAUDI prior · VLN-CE | FID: 18.52 -> 20.20 | Hurt |
| cross-attention instead of FiLM conditioning (ablation) | GAUDI · VLN-CE text conditioning | FID: 20.99 -> 18.50 | Improved |
| cross-attention instead of FiLM conditioning (ablation) | GAUDI · VLN-CE image conditioning | FID: 21.01 -> 19.51 | Improved |
| cross-attention instead of FiLM conditioning (ablation) | GAUDI · VLN-CE categorical conditioning | FID: 18.75 -> 18.74 | No clear effect |
| increase latent noise beta from 0.1 to 1.0 (ablation) | GAUDI latent optimization · VLN-CE reconstruction | PSNR: 39.12 -> 34.82 | Hurt |
| increase latent dimension from 2048 to 8192 (ablation) | GAUDI latent optimization · VLN-CE reconstruction | PSNR: 39.12 -> 36.33 | Hurt |
| condition on environment id (ablation) | GAUDI · VLN-CE per-environment | FID: -50.79 | Improved |
Later papers that compared with this method
1 later accepted paper reports GAUDI as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- EpipolarGAN: Omnidirectional Image Synthesis with Explicit Camera Control (ECCV 2024 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- pi-GAN — Pi-GAN: Periodic Implicit Generative Adversarial Networks for 3D-Aware Image Synthesis (CVPR 2021) · better on 4 of 4 shared results
- GRAF — GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis (NeurIPS 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D Scene Generation: A Survey (IJCV 2026 · similarity 0.89)
- Simple and Effective Synthesis of Indoor 3D Scenes (AAAI 2023 · similarity 0.89)
- GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling (NeurIPS 2024 · similarity 0.88)
- GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representations (ICLR 2020 · similarity 0.88)
- Unconstrained Scene Generation With Locally Conditioned Radiance Fields (ICCV 2021 · similarity 0.88)
- Generative Gaussian Splatting: Generating 3D Scenes with Video Diffusion Priors (ICCV 2025 · similarity 0.88)
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.
