Train & validate

Synthetic Data

Name: Synthetic Battlespace Data Generator
Brand: Runic Defence

Infinite, perfectly-labeled training data — including scenarios you can't safely capture for real.

Generate the rare, the dangerous, and the hard-to-photograph — at scale, fully annotated.

Spec a dataset for your model Explore the system

Synthetic battlespace dataset with labeled vehicles and aircraft across RGB and thermal imaging

Overview

A military-focused synthetic data platform for training and validating AI vision systems. It produces realistic, labeled datasets of vehicles, aircraft, drones, objects, terrain, and operational environments across RGB, thermal, and LWIR imaging modes — building the data you'd otherwise spend years and risk to collect.

It renders physically-based, path-traced 3D scenes in which target models are placed with domain randomization over camera pose, lighting, backgrounds, range, occlusion, weather, and distractors — with every model normalized to a known physical size, so distance and scale are meaningful. Datasets export as COCO (bounding boxes or polygon masks) and KITTI, with automated annotation reports.

A dedicated LWIR pipeline models thermal sensors realistically — automatic gain control, NETD grain, fixed-pattern and column noise, optical blur, vignette, and selectable thermal palettes — and can convert daytime RGB imagery into plausible long-wave-infrared using both physics-based and learned methods.

Military & security use cases

Training models for target detection and classification
Creating rare or difficult-to-capture battlefield scenarios on demand
Generating thermal and LWIR datasets for drone and aircraft detection
Converting existing RGB imagery into realistic thermal datasets
Stress-testing AI against occlusion, distance, clutter, and changing environments
Supporting ISR, surveillance, reconnaissance, and perimeter-defense systems

What it delivers

Scalable, repeatable dataset generation
Faster model development at lower data-collection cost
RGB / thermal / LWIR workflows in one pipeline
Standards-based COCO and KITTI annotations ready for training
Vision systems hardened for real-world operational variability

Key capabilities

Inside Synthetic Data.

RGB, thermal, and LWIR dataset generation
Physically-based, path-traced 3D rendering with multiple quality presets
Domain randomization: pose, lighting, backgrounds, range, occlusion, weather, distractors
Physical size normalization for meaningful range and scale
Realistic LWIR sensor modeling (AGC, NETD grain, fixed-pattern noise, vignette, palettes)
RGB-to-LWIR conversion (procedural physics-based and learned)
COCO (bbox & polygon mask) and KITTI annotation export
Configurable classes-per-frame, distance range, and small-object handling
Automated annotation visualization and PDF reporting

Field view

Path-traced synthetic RGB frame with a drone target rendered against terrain — A path-traced RGB training frame — target placed with randomized pose, lighting, and background.

Synthetic long-wave-infrared thermal frame showing a drone against a cold sky — The same pipeline in LWIR: sensor-realistic thermal imagery for training detectors that must see in the dark.

Product FAQ

Synthetic Data — questions & answers

What data does the generator produce?

Realistic, labeled datasets of vehicles, aircraft, drones, objects, terrain, and operational environments across RGB, thermal, and LWIR imaging modes.

What annotation formats does it export?

Standard formats including COCO and KITTI, with controlled, repeatable scenes and randomized angles, lighting, occlusion, weather, and distractors.

Why use synthetic data?

To create rare or difficult-to-capture scenarios on demand, stress-test models against occlusion, distance, and clutter, and reduce dependency on costly, risky real-world data collection.