README.md

metadata

license: other
tags:
  - svhn
  - street-view-house-numbers
  - image-classification
  - digit-recognition
  - subset
  - computer-vision
source_datasets:
  - ufldl-stanford/svhn
task_categories:
  - image-classification

SVHN (Street View House Numbers) — Subset 🔢

Stratified random subset of SVHN, sourced from ufldl-stanford/svhn.

Splits

Split	Rows	Classes	Per class
train	5,000	10	~500
test	1,000	10	~100
validation	500	10	~50

About SVHN

The Street View House Numbers (SVHN) dataset is a real-world digit recognition dataset obtained from Google Street View images.

• Full train set: 73,257 images
• Full test set: 26,032 images
• Extra set: 531,131 images (not included in this subset)
• Image size: 32 × 32 RGB (cropped_digits config)
• Classes: 10 digits (0–9)
• Config used: cropped_digits — MNIST-like single digit per image

Classes

ID	Digit	ID	Digit
0	0	5	5
1	1	6	6
2	2	7	7
3	3	8	8
4	4	9	9

Usage

from datasets import load_dataset

ds = load_dataset("Chiranjeev007/SVHN-10_Subset")
print(ds)
# DatasetDict({
#   train:      Dataset(num_rows: 5000),
#   validation: Dataset(num_rows: 500),
#   test:       Dataset(num_rows: 1000)
# })

sample = ds["train"][0]
sample["image"]   # PIL Image 32×32 RGB
sample["label"]   # int 0–9

Sampling Details

• Method: Stratified random — proportional per class
• Seed: 42
• Source config: cropped_digits