Feature Launch: Thermal with Depth

Thermal with Depth is now available as a render output.

Maritime perception models earn their keep in the conditions cameras hate: night, fog, glare off the water. Thermal sensors see through all of it - but detection alone isn't enough. Your model also needs to know how far away that vessel is. That's where depth comes in.

Collecting real-world thermal footage with aligned range data is slow and expensive. Now you can render both from the same simulated scene.

What's New

• Thermal with Depth render output: thermal imagery plus a per-pixel depth map from a single render
• One-line setup - switch your render params and you're done
• Both outputs come from the same scene, so thermal and depth are pixel-aligned by construction - no cross-sensor calibration or registration step

Getting started

Change your render params to the new mode:

render_params = RenderParams.ThermalWithDepth()

Every frame now ships with a thermal image and its matching depth map.

Why thermal + depth

Pairing the two modalities unlocks training data that's hard to capture in the real world:

• Multi-modal sensor training. Train and evaluate models that fuse thermal with range information, with ground truth for both.
• Range estimation. Per-pixel depth gives you exact distance labels for every detection in the thermal frame.
• Low-visibility detection. Build datasets for the scenarios that matter most - small targets at night, vessels in fog - with depth ground truth attached.

For the details, see our docs on thermal rendering and depth metadata.

Put it to work

Thermal with Depth is live now. Switch your render params and generate your first multi-modal dataset today.

Working on low-visibility perception? Get in touch