Research map / Detection and segmentation / Open-set and few-shot detection
ACDC: The Adverse Conditions Dataset With Correspondences for Robust Semantic Driving Scene Perception
Christos Sakaridis, Haoran Wang, Ke Li, René Zurbrügg, Arpit Jadon, Wim Abbeloos, Daniel Olmeda Reino, Luc Van Gool, Dengxin Dai
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 SegFormer backbone with MIC instead of DeepLabv2 UDA | UDA segmentation · Cityscapes->ACDC | mIoU: MIC 70.4 vs FDA 45.7 | Improved |
| apply DeepLabv2-based adversarial UDA (AdaptSegNet, ADVENT, BDL) | DeepLabv2 source model · Cityscapes->ACDC | mIoU: 38.0 -> 33.4 (AdaptSegNet), 32.7 (ADVENT) | Hurt |
| add input-level Fourier adaptation (FDA/CISS) | DeepLabv2 source model · Cityscapes->ACDC | mIoU: 38.0 -> 45.7 (FDA), 47.2 (CISS) | Improved |
| add MIC UDA to SegFormer source model | SegFormer · Cityscapes->ACDC | mIoU: 56.8 -> 70.4 | Improved |
| add DAFormer UDA to SegFormer source model | SegFormer · Cityscapes->ACDC | mIoU: 56.8 -> 55.4 | Hurt |
| add weak supervision via image correspondences (MGCDA) | RefineNet · Cityscapes->ACDC | mIoU: 43.7 -> 48.7 | Improved |
| add target-domain adaptation (CISS) on sensor-level shift | DeepLabv2 · Cityscapes->ACDC-Reference | mIoU: 51.7 -> 63.5 vs oracle 62.5 | Improved |
| add MIC adaptation on sensor-level shift | SegFormer · Cityscapes->ACDC-Reference | mIoU: 68.4 -> 77.0 vs oracle 72.4 | Improved |
| add UDA to detectors (adversarial DA-Faster, SADA) | Faster R-CNN · Cityscapes->ACDC | AP50 box: 20.8 -> 19.8 (DA-Faster), 19.7 (SADA) | Hurt |
| add MIC (SADA) to Faster R-CNN | Faster R-CNN · Cityscapes->ACDC | AP50 box: 20.8 -> 25.9 | Improved |
| train uber model on all conditions instead of condition expert (ablation) | HRNet · ACDC all conditions | mIoU: 70.9 -> 75.0 | Improved |
| train uber model on all conditions instead of fog expert (ablation) | Mask R-CNN · ACDC fog | AP mask: 15.6 -> 24.4 | Improved |
Closest papers by meaning
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · similarity 0.85)
- Abductive Ego-View Accident Video Understanding for Safe Driving Perception (CVPR 2024 · similarity 0.85)
- Parsing All Adverse Scenes: Severity-Aware Semantic Segmentation with Mask-Enhanced Cross-Domain Consistency (AAAI 2024 · similarity 0.85)
- ZOPP: A Framework of Zero-shot Offboard Panoptic Perception for Autonomous Driving (NeurIPS 2024 · similarity 0.85)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.85)
- AIDE: An Automatic Data Engine for Object Detection in Autonomous Driving (CVPR 2024 · similarity 0.85)
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.
