Whilst machine learning can be a complex process, our process of applying it in your business is simple.
Take a look at our 6-step process of how we take your photos of assets and defects and turn them into the millions of synthetic photos needed to train the model.
Our secret sauce is Morpheus, the simulator that generates the photos.
We work closely with our client to define the problem and then we put our software to work. Here’s the step by step process of how we work with our clients:
- You nominate the asset to be inspected eg a power pole
- You nominate the defect on that asset you’d like to inspect eg rust, cracks, broken wires etc
- You supply us with a handful of photos (we can work with as few as 10 photos) of the asset and the defective asset and we create 3D models of each (‘digital twins’)
- We feed that data into Morpheus – our simulator (our secret sauce)
- The simulator replicates that asset in all sorts of conditions (eg different backgrounds, environments, lighting, shading, colours, textures etc) so it can generate high resolution, photo-realistic synthetic photos
- We now have the millions of ‘synthetic’ photos of the asset and the defective asset, and the volume of data we need to train the model
- Our client runs their ‘real’ images of their assets eg power pole (sourced by drones, helicopters, manual inspections etc) through the training model we have just created
- When the model ‘sees’ a defect in the photos of those assets, it flags that photo for closer inspection by the human assessment team, who prioritise it and schedule it for maintenance and repair
All the synthetic photos are automatically labelled
Best of all, our synthetic photos are automatically labelled, which means the costly step of manually labelling each and every photo is eliminated. Most photos cost up to 84 cents to be labelled. If you need over a million photos labelled, that’s a cost of $840,000 instantly eliminated.
Our software provides you with the missing piece of the Machine Learning puzzle. Our synthetic photos makes machine learning affordable and accessible for all.
How many photos do we need to start work?
We know sourcing photos of assets and defects can be difficult. Some of our clients often only have one or two images of hard-to-find defects. We can work with as few as 10 photos or less of your asset and defect.
What sort of photos can we work with?
The photos could be from a:
- Mobile phone
- CAD drawing
- Product catalogue
- Video camera