anywaylabs/synthetic-suspended-load-detection

Synthetic Suspended Load Detection Dataset by AnywayLabs.ai

Dataset Summary

This dataset contains fully synthetic images for suspended load detection in factory and construction environments, generated by AnywayLabs.ai.

The dataset is designed for supervised object detection of workers and suspended loads, covering multiple camera viewpoints (FOV). It targets real-world workplace safety monitoring and is intended to train models that detect dangerous proximity between workers and overhead suspended loads.

Key Features

• Fully synthetic suspended load detection dataset
• Supervised object detection formulation
• YOLO-format annotations
• Multiple camera fields-of-view (FOV1–FOV10)
• 2 object classes
• 500 annotated training images

---

Sample Images

FOV1	FOV2	FOV3

Dataset Structure

Format

• Task: Object Detection (Suspended Load / Workplace Safety Detection)
• Annotation Format: YOLO (normalized bounding boxes)
• Image Resolution: 1376 × 768 (primary)

Directory Structure

images/
  train/
labels/
  train/
data.yaml

---

Classes

ID	Class Name	Description
0	worker	Worker present in the scene
1	suspended_load	Overhead suspended load (crane hook, cargo, etc.)

---

Data Generation

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

Capabilities

• Behavior-level control:

  • Worker pose and position variation
  • Proximity to suspended load
  • Occlusion and partial visibility

• Scene-level control:

  • Lighting conditions
  • Industrial and construction background variation
  • Camera placement and field-of-view

• Structural consistency:

  • Realistic load rigging and crane geometry
  • Consistent worker body geometry

Design Philosophy

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

• Expand the behavioral distribution beyond limited real industrial samples
• Introduce controlled variation across load positions and worker proximity
• Improve robustness to unseen real-world factory and construction scenarios

---

Annotation Process

• Bounding boxes are generated for each object instance

• Labels follow YOLO format:

  class_id x_center 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

• Industrial safety monitoring model training
• Synthetic pretraining before fine-tuning on real factory/construction data
• Safety system development for factories and construction sites
• Multi-view suspended load proximity detection research

Compatible Frameworks

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

---

Comparison with Real-World Safety Datasets

Feature	Real-World Safety Datasets	This Dataset
Data Source	Real CCTV / sensor recordings	Fully synthetic
Privacy Concerns	High (person identity)	None
Annotation Effort	Manual, costly	Automated
Scenario Coverage	Limited by recording	Configurable
Class Balance	Often imbalanced	Controllable
Training Paradigm	Supervised	Fully supervised

---

Citation

If you use this dataset, please cite:

Synthetic Suspended Load Detection Dataset – AnywayLabs.ai