> ML_DATASET // STREET-VIEW-HOUSE-NUMBERS_v1.0
The Street View House Numbers (SVHN)
Stanford University / Google (Netzer et al.) · Real-World Optical Character Recognition · 600,000+ 32x32 color digit images cropped from Google Street View house numbers
Real-World Optical Character RecognitionNon-commercial research only600,000+ 32x32 color digit images cropped from Google Street View house numbersopen
Dataset Profile & Characteristics
Label Type:Categorical digit (0 to 9)
Languages:en
License Tier:non-commercial
Modalities:image
Intended Use
- Benchmarking robustness of vision classifiers on natural scene character recognition
Prohibited / Discouraged Use
- Automated property geolocation surveillance
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
House numbers on residential buildings; zero human faces or PII.
Known Bias:
Real-world lighting variation, distracting neighboring digits, motion blur, and low contrast.
Known Benchmark Leakage:
Clean train, test, and extra training sets.