Research map / 3D vision / 3D generation
LSD-3D: Large-Scale 3D Driving Scene Generation with Geometry Grounding
Julian Ost, Andrea Ramazzina, Amogh Joshi, Maximilian Bömer, Mario Bijelic, Felix Heide
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 |
|---|---|---|---|
| replace GGDS with latent-space SDS (ablation) | LSD-3D · Generated scenes | FID: 155.87 -> 326.68 | Hurt |
| add DDIM inversion instead of random noise sampling (ablation) | GGDS · Generated scenes | FID: 455.93 -> 155.87 | Improved |
| use multi-step denoising (N=5) instead of single step (ablation) | GGDS · Generated scenes | FID: 218.91 -> 155.87 | Improved |
| add geometry-grounding diffusion guidance (ablation) | GGDS · Generated scenes | FID: 175.25 -> 155.87 | Improved |
| generate explicit 3D Gaussian scene instead of video + 2DGS (novel views) | LSD-3D vs Vista+2DGS · Waymo novel views | FVD: 1190.76 -> 974.36 | Improved |
| generate explicit 3D Gaussian scene instead of video + 2DGS (novel views) | LSD-3D vs GEN3C · Waymo novel views | FID: 273.27 -> 119.18 | Improved |
| generate explicit 3D scene on seen views | LSD-3D vs GEN3C · Waymo seen views | FID: 97.71 vs 119.38 | Hurt |
| 3D grounding does not reduce prompt adherence | LSD-3D vs Vista+2DGS · Prompt adherence | CLIP score: 26.51 vs 26.03 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GEN3C — GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025) · better on 4 of 7 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
- SEM-ROVER: Semantic Voxel-Guided Diffusion for Large-Scale Driving Scene Generation (ECCV 2026 · similarity 0.89)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · similarity 0.88)
- UniScene: Unified Occupancy-centric Driving Scene Generation (CVPR 2025 · similarity 0.87)
- GenAssets: Generating in-the-wild 3D Assets in Latent Space (CVPR 2025 · similarity 0.87)
- RLGF: Reinforcement Learning with Geometric Feedback for Autonomous Driving Video Generation (NeurIPS 2025 · similarity 0.87)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · 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.
