> ML_DATASET // VISUAL-GENOME-DATASET_v1.0
Visual Genome: Connecting Language and Vision (Krishna et al. 2017)
Stanford University (Ranjay Krishna et al.) · Visual Relational Knowledge & Scene Graphs · 108,077 images with 5.4M region descriptions, 2.3M visual relationships, and 2.8M attributes
Visual Relational Knowledge & Scene GraphsCC-BY-4.0108,077 images with 5.4M region descriptions, 2.3M visual relationships, and 2.8M attributesopen
Dataset Profile & Characteristics
Label Type:Scene graphs: localized bounding boxes, attributes, and directed relational predicate triples
Languages:en
License Tier:permissive-open-source
Modalities:image, text, graph
Intended Use
- Benchmarking scene graph generation, visual relationship detection, and grounded VQA
Prohibited / Discouraged Use
- Facial recognition
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
Photographic images with human activities in public and private spaces.
Known Bias:
Heavy visual relationship long-tail (e.g., "on", "has", "wearing" dominate).
Known Benchmark Leakage:
Clean split mapped to MS COCO image identifiers.