---
title: "Crowd Counting Dataset"
description: "647 photos 1000-13,000 people in a crowd Keypoint detection"
url: "https://unidata.pro/datasets/people-detection-image/"
date_modified: "2025-10-01T07:46:50+03:00"
language: "en-US"
---
Crowd counting dataset is a labeled dataset including images of dense crowds with key point annotations and metadata on crowd sizes, designed for training deep learning models in crowd counting, object detection, and data analysis across a larger scale of counting datasets

## Dataset Structure

### The Numbers Section

**Numbers list:**

- **Number:** 647 — **Text:** Photos
- **Number:** 1,000 - 13,000 — **Text:** People in a crowd

### Tooltip Section

**Tooltip items:**

- **Name:** Smart City
- **Name:** Computer Vision
- **Name:** Machine Learning
- **Name:** Keypoint Detection
- **Name:** Security

### Dataset Info

**Table with data:**

| Characteristic | Data |
| --- | --- |
| Description | Crowd photos with labeling for determining crowd density and movement, and counting the number of people |
| Data types | Image |
| Tasks | Crowd control, object detection, and computer vision |
| Total number of photos | 647 |
| Type of crowd density | 0-1000, 1000-2000, 2000-3000, 3000-4000, 4000-5000 |
| Labeling | Metadata (type of crowd density) and key points labeling |

**Media Slider:**

- **Image in the slider:** ![](https://unidata.pro/wp-content/uploads/2024/06/185.webp)
- **Image in the slider:** ![](https://unidata.pro/wp-content/uploads/2024/06/186.webp)

**Link to the sample:** [Download sample](https://drive.google.com/drive/folders/1If1gGUNvUDvtRssGXH-_9-5nuIu1Gae8?usp=drive_link)

### Technical  characteristics

**Table with data:**

| Characteristic | Data |
| --- | --- |
| Image Extensions | JPG |
| Extension of labeling file | JSON |

**Source and collection methodology:** Source and collection methodology:  Data was obtained by parsing photos of protests, concerts, and other mass events from the internet.

### Statistics - Charts

**Charts with Titles:** - **Shortcode:** [ays_chart id='41'] — **caption above the graph:** Number of images by the type of a crowd:

### Dataset Use Cases - слайдер

**Industry Cards:**

- **Industry:** Smart City Management — **Title:** Urban Planning and Public Safety — **Text:** Crowd Counting Dataset can help city planners and safety agencies analyze crowd sizes in dense public spaces. By using counting algorithms and object detection, authorities can monitor pedestrian traffic, detect large-scale gatherings, and optimize resource allocation, creating safer and more efficient urban environments through data analysis and deep learning models.
- **Industry:** Retail & Commercial Spaces — **Title:** Customer Flow and Store Analytics — **Text:** For retailers, this crowd-labeled dataset can be useful for measuring customer movement in malls, supermarkets, and events. By applying crowding counting methods, businesses can optimize store layouts, reduce congestion, and improve the shopping experience. The datasets include images of varied crowd sizes, making them ideal for building AI-driven counting systems for real-world retail analysis.
- **Industry:** Event Management & Security — **Title:** Monitoring Dense Crowds in Real Time — **Text:** Event organizers and security teams rely on the human crowd dataset to track dense crowds and ensure safety. With counting datasets that include diverse crowd scenarios, AI models learn to detect the largest number of people quickly, helping reduce risks, improve surveillance, and enhance decision-making during concerts, sports, or festivals.
- **Industry:** Transportation & Mobility — **Title:** Crowd Flow in Transit Hubs — **Text:** The crowd dataset supports analysis of larger-scale gatherings in airports, train stations, and bus terminals. By training deep learning models on these annotated images, transport authorities can predict congestion, optimize passenger movement, and implement crowd counting systems for smoother operations in high-density transportation networks and urban mobility planning.

### Fact

**FAQs Heading:** FAQs

**List of Questions:**

- **Question:** What types of annotations are provided? — **Answer:** The dataset provides metadata about crowd density categories as well as key point labeling for individuals in images.
- **Question:** How was the data collected? — **Answer:** Crowd Counting Dataset was created by parsing images from the internet, including protests, concerts, and other large-scale gatherings. This approach ensures a wide range of crowd sizes and environments, giving researchers diverse training data for crowd analysis and recognition technology.
- **Question:** How long does it take to receive the dataset? — **Answer:** The process begins once you submit a request. Our team will review the details with you and complete the necessary documents. After signing and payment, the dataset is delivered within 3–10 days.
- **Question:** Is this a real-world dataset or synthetic data? — **Answer:** Crowd Counting Dataset is a real-world dataset, not synthetic data. It contains photos of actual crowds taken from events such as protests, concerts, and other mass gatherings.
- **Question:** What crowd density range does the dataset cover? — **Answer:** Images are categorized into five density bands: 0–1,000, 1,000–2,000, 2,000–3,000, 3,000–4,000, and 4,000–5,000 people. This banding lets models train on a graduated scale from sparse gatherings to dense crowds.
- **Question:** Do Unidata datasets follow GDPR or other data privacy regulations? — **Answer:** Yes. Each dataset is fully GDPR-compliant and meets data protection obligations. All collected information comes from permissible sources.
- **Question:** What is key point annotation, and why does it matter for crowd counting? — **Answer:** Key point annotation marks the location of each individual within a crowd image, rather than using bounding boxes. This precision is critical for accurately estimating headcounts in dense, overlapping scenes.
- **Question:** How are Unidata datasets licensed? — **Answer:** Unidata datasets run on a dual licensing model: free evaluation samples are provided, while the entire dataset must be purchased.

[Full list of this site's AI-readable pages](https://unidata.pro/llms.txt)