Research map / Point clouds / Segmentation and detection
Toward Generating Realistic 3D Semantic Training Data for Autonomous Driving
Lucas Nunes, Rodrigo Marcuzzi, Jens Behley, Cyrill Stachniss
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 single sparse 3D latent diffusion instead of XCube multi-resolution at 0.1 m | 3D scene generator · SemanticKITTI generated scenes (0.1 m) | MMD: 0.101 -> 0.073 | Improved |
| use single sparse 3D latent diffusion instead of XCube at 0.2 m | 3D scene generator · SemanticKITTI generated scenes (0.2 m) | MMD: 0.160 -> 0.146 | Improved |
| use single-VAE method instead of XCube for realism measured by real-trained segmenter at 0.1 m | 3D scene generator · mIoU on generated scenes (0.1 m) | mIoU: 27.24 -> 53.09 | Improved |
| use single-VAE method instead of XCube at 0.2 m | 3D scene generator · mIoU on generated scenes (0.2 m) | mIoU: 40.02 -> 50.83 | Improved |
| increase resolution from 0.2 m to 0.1 m | Ours · mIoU on generated scenes | mIoU: 50.83 -> 53.09 | Improved |
| increase resolution from 0.2 m to 0.1 m | XCube · mIoU on generated scenes | mIoU: 40.02 -> 27.24 | Hurt |
| complement 10% real data with our synthetic scenes (LiDAR-simulated) | segmentation network · SemanticKITTI val, LiDAR-simulated | mIoU: 45.58 -> 55.54 | Improved |
| complement 10% real data with XCube synthetic scenes (dense) | segmentation network · SemanticKITTI val, dense | mIoU: 56.22 -> 53.75 | Hurt |
| complement 10% real data with our synthetic scenes (dense) | segmentation network · SemanticKITTI val, dense | mIoU: 56.22 -> 61.73 | Improved |
| complement 10% real data with XCube synthetic scenes (LiDAR-simulated) | segmentation network · SemanticKITTI val, LiDAR-simulated | mIoU: 45.58 -> 49.28 | Improved |
| add 75% extra synthetic scenes to full real training set (dense) | segmentation network · SemanticKITTI val, dense | mIoU: 61.08 -> 64.14 | Improved |
| add 75% extra synthetic scenes to full real training set (LiDAR-simulated) | segmentation network · SemanticKITTI val, LiDAR-simulated | mIoU: 55.59 -> 57.77 | Improved |
Closest papers by meaning
- Cross-Dataset Collaborative Learning for Semantic Segmentation in Autonomous Driving (AAAI 2022 · similarity 0.89)
- TRoVE: Transforming Road Scene Datasets into Photorealistic Virtual Environments (ECCV 2022 · similarity 0.89)
- 3D Annotation-Free Learning by Distilling 2D Open-Vocabulary Segmentation Models for Autonomous Driving (AAAI 2025 · similarity 0.88)
- DriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation (CVPR 2025 · similarity 0.87)
- SemCity: Semantic Scene Generation with Triplane Diffusion (CVPR 2024 · similarity 0.87)
- UniScene: Unified Occupancy-centric Driving Scene Generation (CVPR 2025 · similarity 0.86)
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.
