Introducing the Simuletic Synthetic CCTV Weapon-Detection Dataset

Advancing AI Safety With Realistic Synthetic Data

Today we're releasing Simuletic's first open-source synthetic dataset for AI powered weapon detection software — a YOLO-annotated CCTV weapon detection dataset built to train models on the scenarios real footage almost never covers, including AI concealed weapons detection where a handgun is partially hidden behind a jacket, bag, or the person's own body.

Modern computer vision models are extremely capable — but they still struggle with rare, high-risk edge cases like weapons appearing in real-world surveillance footage. These events are:

• rare (thankfully)
• highly sensitive ethically and legally
• difficult to collect responsibly
• almost impossible to annotate at scale
• often constrained by privacy & regulatory requirements

Synthetic data changes this.

This release is part of our mission at Simuletic to make AI safer and more reliable by providing diverse, ethical, and controllable datasets tailored for real-world safety scenarios.

Why Synthetic CCTV Weapon Data?

Traditional datasets face major challenges:

Our goal is to support security, research, and surveillance AI teams building responsible detection systems, without compromising privacy or sourcing sensitive footage.

Dataset Overview

This dataset is a sample produced by Simuletic's synthetic data pipeline, built for high-variance CCTV edge-case simulation. Larger commercial datasets & custom scenarios are available.

Data Structure

dataset/
 ├─ images/
 │   └─ *.jpg / *.png
 ├─ labels/
 │   └─ *.txt   # YOLO format
 ├─ annotations.csv (optional)
 └─ dataset.yaml

Example YOLO Label Line

0 0.45 0.55 0.20 0.30   # person
1 0.60 0.70 0.10 0.15   # weapon

Sample dataset.yaml

path: /path/to/dataset
train: images
val: images
names:
  0: person
  1: weapon

Download & Explore

If you train a model on this dataset, we'd love to see your results!

Ethics, Safety & Limitations

We believe strongly in ethical synthetic data for public safety.

We plan to expand the diversity of:

• environments
• subjects and clothing
• behavior patterns
• camera viewpoints
• lighting and weather conditions

Always audit, stress-test, and validate your model before real deployment.