> ML_BENCHMARK // CITYSCAPES-SEMANTIC-SEGMENTATION-MIOU_v1.0
Cityscapes Semantic Segmentation (mIoU)
Urban Semantic Segmentation · task-image-segmentation · active-headroom
Urban Semantic Segmentationactive-headroomhigh-vram-dependent
Evaluation Protocol
500 validation images evaluated on 19 urban driving classes at full 2048x1024 resolution.
Baseline & Metrics
Canonical Baseline:FCN-8s: 65.3% | DeepLabv3+: 81.3% | SegNeXt: 83.9%
Evaluated Metrics:
mean Intersection-over-Union (mIoU %)iIoU
Contamination & Leakage Risks
Pretraining on ADE20K or ImageNet-21k is standard; city splits prevent geographic leak.
Reproducibility Concerns
Single-scale vs multi-scale with flip inference switches.
CONNECTED DATASETCityscapes: Semantic Understanding of Urban Street ScenesAutonomous Driving & Urban Perception · 5,000 fine-annotated stereo images across 50 cities, 20,000 coarsely annotated frames
View Dataset Spec