anywaylabs/synthetic-jewellery-theft-detection

Synthetic Jewellery Theft Detection Dataset by AnywayLabs.ai

Need a custom synthetic dataset for your own theft detection use case?

This dataset is an open-source sample of our synthetic data generation work at AnywayLabs.

If you're working on:

• —industrial defect detection
• —visual inspection
• —supervised anomaly detection
• —hard-to-collect defect classes
• —synthetic data for computer vision training

You can request a custom synthetic dataset here, or email: contact@anywaylabs.ai

Dataset Summary

This dataset contains fully synthetic images for jewellery theft detection in retail and surveillance scenarios, generated by AnywayLabs.ai.

The dataset is designed for supervised object detection of theft-related behaviors, covering three action classes across multiple camera viewpoints (FOV). It targets real-world retail security monitoring and is intended to train models that detect shoplifting and theft events in jewellery stores and similar environments.

Key Features

• —Fully synthetic jewellery theft detection dataset
• —Supervised object detection formulation
• —YOLO-format annotations
• —Multiple camera fields-of-view (FOV1, FOV2, FOV3 and jewelry-specific views)
• —3 theft behavior classes
• —571 annotated training images

Sample Images

<table> <tr> <th>FOV1</th> <th>FOV2</th> <th>FOV3</th> </tr> <tr> <td><img src="images/train/6.png" width="300"/></td> <td><img src="images/train/2.png" width="300"/></td> <td><img src="images/train/0.png" width="300"/></td> </tr> </table>

Dataset Structure

Format

• —Task: Object Detection (Theft / Shoplifting Detection)
• —Annotation Format: YOLO (normalized bounding boxes)
• —Image Resolution: 1408 × 768 (primary), 1376 × 768, 1200 × 896

Directory Structure

images/ train/ labels/ train/ data.yaml

Classes

Data Generation

The dataset is generated using AnywayLabs.ai's proprietary synthetic dataset generation framework.

Capabilities

• —Behavior-level control:
• —Action type (e.g., stealing, picking, carrying bag)
• —Pose variation and body positioning
• —Occlusion and partial visibility

• —Scene-level control:
• —Lighting conditions
• —Retail interior background variation
• —Camera placement and field-of-view

• —Structural consistency:
• —Subject body geometry preserved
• —Realistic action integration

Design Philosophy

Instead of replicating real data exactly, the dataset is designed to:

• —Expand the behavioral distribution beyond limited real surveillance samples
• —Introduce controlled variation across theft action appearance
• —Improve robustness to unseen real-world shoplifting behaviors

Annotation Process

• —Bounding boxes are generated for each behavior instance
• —Labels follow YOLO format:

classid xcenter y_center width height

• —Bounding box tightness is manually reviewed
• —Annotation consistency is maintained across all camera views

Limitations

• —Validation sets are not included in the public release
• —Performance may vary across camera placements and lighting conditions
• —Bounding boxes do not capture fine-grained body pose (compared to keypoint annotations)

Usage

Recommended Use Cases

• —Retail shoplifting detection model training
• —Synthetic pretraining before fine-tuning on real surveillance data
• —Security system development for jewellery and high-value retail stores
• —Multi-view theft behavior recognition research

Compatible Frameworks

• —Ultralytics YOLO (recommended)
• —Detectron2 (after conversion)
• —Any YOLO-compatible pipeline

Comparison with Real-World Theft Detection Datasets

Need a custom synthetic dataset for your own theft detection use case?

This dataset is an open-source sample of our synthetic data generation work at AnywayLabs.

If you're working on:

• —industrial defect detection
• —visual inspection
• —supervised anomaly detection
• —hard-to-collect defect classes
• —synthetic data for computer vision training

You can request a custom synthetic dataset here, or email: contact@anywaylabs.ai

Citation

If you use this dataset, please cite:

Synthetic Jewellery Theft Detection Dataset – AnywayLabs.ai