How expensive is it to compute?

The first cloud run built 597k Gaussians from 88 photos for $0.75 of GPU time, debugging included.

Can it run on our own hardware?

The light steps can. Structure from motion and splat training need a GPU, rented in the cloud when there is none locally.

Service

Digital twins from drone video, and training data made inside them

Detection models for drones need thousands of labelled aerial images of the places they will fly over. I rebuild the real place once from drone footage and generate labelled data inside it, using only commercially usable licences.

Discuss a project →

What you get

A survey plan: height, ground resolution, orbits and overlap
Camera poses aligned to GPS and a photorealistic Gaussian-splat scene
A collision mesh and the twin in Isaac Sim with PX4 flights
Labelled data in COCO and YOLO formats for training detectors

Where I have already done this

R&D · drones and synthetic dataReal2simDrone video becomes a photorealistic 3D copy of the place, where drones can be trained.597k Gaussians in the first cloud splat$0.75 for that run, debugging included

Questions

How expensive is it to compute?: The first cloud run built 597k Gaussians from 88 photos for $0.75 of GPU time, debugging included.
Can it run on our own hardware?: The light steps can. Structure from motion and splat training need a GPU, rented in the cloud when there is none locally.

Digital twins from drone video, and training data made inside them

What you get

Where I have already done this

Questions

Describe your task