Research map / Point clouds / Segmentation and detection
Drive&Segment: Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation
Antonin Vobecky, David Hurych, Oriane Siméoni, Spyros Gidaris, Andrei Bursuc, Patrick Pérez, Josef Sivic
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 |
|---|---|---|---|
| add LiDAR-based segment extraction instead of image-based FH segments (ablation) | D&S Segmenter · Cityscapes | mIoU: 15.8 -> 20.4 (+4.6) | Improved |
| add LiDAR-based segment extraction instead of image-based FH segments (ablation) | D&S RN18+FPN · Cityscapes | mIoU: 15.5 -> 17.4 (+1.9) | Improved |
| add teacher-student distillation (student without LiDAR constraints) (ablation) | D&S Segmenter · Cityscapes | mIoU: 20.4 -> 20.8 (+0.4) | Improved |
| add LiDAR spatial constraints to distillation (ablation) | D&S Segmenter · Cityscapes | mIoU: 20.8 -> 21.8 (+1.4) | Improved |
| add teacher-student distillation (ablation) | D&S RN18+FPN · Cityscapes | mIoU: 17.4 -> 18.8 (+1.4) | Improved |
| add LiDAR spatial constraints to distillation (ablation) | D&S RN18+FPN · Cityscapes | mIoU: 18.8 -> 19.5 | Improved |
| add distillation to PiCIE (ablation) | PiCIE RN18+FPN · Cityscapes | mIoU: 13.7 -> 14.8 (+1.1) | Improved |
| add LiDAR constraints to PiCIE distillation (ablation) | PiCIE RN18+FPN · Cityscapes | mIoU: 14.8 -> 15.1 | Improved |
| use DeiT features instead of DINO for k-means labeling (ablation) | D&S Segmenter · Cityscapes | mIoU: 20.2 -> 21.7 | Improved |
| use supervised ResNet18 features instead of DINO (ablation) | D&S Segmenter · Cityscapes | mIoU: 20.2 -> 19.6 | Hurt |
| vary k-means and network initialization seeds (ablation) | D&S Segmenter · Cityscapes | mIoU: 20.4 ± 0.5 | No clear effect |
| replace supervised ImageNet init with DINO self-supervised ViT-S Segmenter | D&S · Cityscapes 19 classes | mIoU: 19.5 -> 21.8 | Improved |
Closest papers by meaning
- Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation (IJCV 2025 · similarity 0.99)
- A Curriculum Domain Adaptation Approach to the Semantic Segmentation of Urban Scenes (TPAMI 2019 · similarity 0.90)
- Exploring Scene Affinity for Semi-Supervised LiDAR Semantic Segmentation (CVPR 2025 · similarity 0.88)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · similarity 0.88)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · similarity 0.87)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (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.
