Research map / 3D vision / 3D generation
Bolt3D: Generating 3D Scenes in Seconds
Stanislaw Szymanowicz, Jason Y. Zhang, Pratul Srinivasan, Ruiqi Gao, Arthur Brussee, Aleksander Holynski, Ricardo Martin-Brualla, Jonathan T. Barron, Philipp Henzler
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 |
|---|---|---|---|
| use generative latent diffusion (Bolt3D) instead of regression | Flash3D · RE10K 1-view | PSNR: 17.40 -> 21.03 | Improved |
| use generative approach instead of DepthSplat regression | DepthSplat · DL3DV 4-view | PSNR: 19.48 -> 20.64 | Improved |
| use Bolt3D instead of optimization-based CAT3D | CAT3D · RE10K 3-view | PSNR: 29.56 -> 27.00 | Hurt |
| use Bolt3D instead of optimization-based CAT3D (cost) | CAT3D · RE10K 3-view | GPU-min: 77.28 -> 0.25 | Improved |
| use Bolt3D instead of LatentSplat (adversarial VAE-GAN) | LatentSplat · RE10K 2-view | FID: 2.79 -> 3.66 | Hurt |
| train geometry encoder rather than freezing pre-trained one (ablation) | Geometry VAE · Pointmap autoencoding | Rel.: 1.63 -> 0.99 | Improved |
| add rec loss re-weighting on distant points (ablation) | Geometry VAE · Pointmap autoencoding | Rel.: 1.68 -> 0.99 | Improved |
| add gradient loss (ablation) | Geometry VAE · Pointmap autoencoding | Rel.: 1.16 -> 0.99 | Improved |
| use geometry-trained autoencoder instead of pre-trained image autoencoder (ablation) | Geometry AE · Real/unbounded pointmaps | Rel.x100: 17.9 -> 0.670 | Improved |
| use ViT decoder instead of conv decoder (ablation) | Geometry AE · Pointmaps real data | delta 1.01 / reprojection: 81.4 -> 81.5 | No clear effect |
| add cross-attention in Gaussian head (ablation) | Gaussian head · Appearance ablation | PSNR: 23.80 -> 24.72 | Improved |
| add Gaussian head (ablation) | Bolt3D · Appearance ablation | PSNR: 21.94 -> 24.72 | Improved |
Later papers that compared with this method
1 later accepted paper reports Bolt3D as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Lyra: Generative 3D Scene Reconstruction via Video Diffusion Model Self-Distillation (ICLR 2026 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- CAT3D — CAT3D: Create Anything in 3D with Multi-View Diffusion Models (NeurIPS 2024) · better on 1 of 6 shared results
- LatentSplat — latentSplat: Autoencoding Variational Gaussians for Fast Generalizable 3D Reconstruction (ECCV 2024) · better on 3 of 3 shared results
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 2 of 2 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
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.90)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.90)
- Prometheus: 3D-Aware Latent Diffusion Models for Feed-Forward Text-to-3D Scene Generation (CVPR 2025 · similarity 0.90)
- DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior (ICLR 2024 · similarity 0.89)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.89)
- Wonderland: Navigating 3D Scenes from a Single Image (CVPR 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.
