Research map / Detection and segmentation / 3D detection and driving
Cross-Dataset Collaborative Learning for Semantic Segmentation in Autonomous Driving
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train one unified model jointly across datasets using CDCL | PSPNet (ResNet-18) · Cityscapes, BDD100K, and CamVid validation sets (three-dataset setting) | mIoU: +5.65%, +6.57%, and +5.79% | Helped |
| Add dataset alternation training to shared Conv and separate BN (ablation) | CDCL with PSPNet (ResNet-18) · Cityscapes val (Cityscapes + BDD100K) | mIoU: 68.55% -> 72.63% | Helped |
| Add dataset alternation training to shared Conv and separate BN (ablation) | CDCL with PSPNet (ResNet-18) · BDD100K val (Cityscapes + BDD100K) | mIoU: 58.93% -> 60.47% | Helped |
| Use separate BN layers for each dataset rather than shared BN (ablation) | Joint training with shared Conv · Cityscapes val (Cityscapes + BDD100K) | mIoU: 62.10% -> 68.55% | Helped |
| Use separate BN layers for each dataset rather than shared BN (ablation) | Joint training with shared Conv · BDD100K val (Cityscapes + BDD100K) | mIoU: 53.34% -> 58.93% | Helped |
| Share Conv weights across datasets while keeping BN unshared (ablation) | Joint training without DAT · Cityscapes val (Cityscapes + BDD100K) | mIoU: 67.75% -> 68.55% | Helped |
| Share Conv weights across datasets while keeping BN unshared (ablation) | Joint training without DAT · BDD100K val (Cityscapes + BDD100K) | mIoU: 53.88% -> 58.93% | Helped |
| Share BN across datasets instead of using dataset-specific BN (ablation) | Joint training with shared Conv · Cityscapes val (Cityscapes + BDD100K) | mIoU: 68.55% -> 62.10% | Hurt |
| Share BN across datasets instead of using dataset-specific BN (ablation) | Joint training with shared Conv · BDD100K val (Cityscapes + BDD100K) | mIoU: 58.93% -> 53.34% | Hurt |
| Update Conv during sequential Cityscapes-to-BDD100K adaptation | Two-stage domain adaptation · Cityscapes val | mIoU: 67.52% -> 1.98% | Hurt |
| Use CDCL rather than the single-dataset model | SalsaNext · SemanticKITTI and nuScenes validation sets | mIoU: 1.0% and 0.3% | Helped |
Closest papers by meaning
- Importance-Aware Semantic Segmentation in Self-Driving with Discrete Wasserstein Training (AAAI 2020 · similarity 0.89)
- PanDA: Unsupervised Domain Adaptation for Multimodal 3D Panoptic Segmentation in Autonomous Driving (CVPR 2026 · similarity 0.88)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · similarity 0.88)
- RMP-SAM: Towards Real-Time Multi-Purpose Segment Anything (ICLR 2025 · similarity 0.87)
- UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · similarity 0.87)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · 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.