SI-Lab/Stepcount-500

---
dataset_info:
features:
- name: images
sequence: image
- name: problem
dtype: string
- name: answer
dtype: string
splits:
- name: train
num_bytes: 106400000
num_examples: 500
download_size: 106400000
dataset_size: 106400000
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# Stepcount-500

## Dataset Description

Stepcount-500 is a benchmark dataset of 500 counting samples with object counts ranging from 11 to 50.

## Dataset Structure

### Data Fields

- `images`: A sequence of images (one image per sample)
- `problem`: The question text describing the counting task
- `answer`: The ground truth count number

### Data Splits

- **train**: 500 evaluation samples

## Dataset Statistics

- Count range: 11-50
- Distribution: 11-20 (130), 21-30 (130), 31-40 (120), 41-50 (120)
- Total dataset size: approximately 106 MB

## Usage

```python
from datasets import load_dataset
ds = load_dataset("SI-Lab/Stepcount-500")
```