> ML_DATASET // ADE20K-SCENE-PARSING_v1.0
ADE20K Scene Parsing Benchmark
MIT CSAIL (Zhou et al.) · Computer Vision & Scene Understanding · 22,210 densely annotated images spanning 150 semantic object and stuff categories
Computer Vision & Scene UnderstandingBSD-3-Clause (Annotations)22,210 densely annotated images spanning 150 semantic object and stuff categoriesopen
Dataset Profile & Characteristics
Label Type:Dense pixel-level semantic masks and object part segmentations
Languages:en
License Tier:permissive-open-source
Modalities:image
Intended Use
- Evaluating dense scene parsing and general-purpose semantic segmentation backbones (Swin, ViT)
Prohibited / Discouraged Use
- Real-time embedded perception requiring low class counts
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
Everyday indoor and outdoor photographic scenes from SUN and Places databases.
Known Bias:
Complex cluttered scenes with long-tailed semantic category distribution.
Known Benchmark Leakage:
Official benchmark partitions maintained with blinded test server.