Our technology initiates a chain reaction:
Immediate availability of datasets for an endless scope of environments and objects, and all types and models of sensors
Accelerated and cost-effective model training
Quick adoption to new objects and expansion to new environments
Discovery of any object, from common to rare, in remote sensing imagery
Here’s how we create 3D models to produce virtual, synthetic images that replace real images.
Creation of the actual environment - urban, maritime, agricultural, etc. A randomization factor for size and relationship between elements is applied.
Placing objects-of-interest, already annotated, based on requirements. Adding annotated ‘distractors’ - similar to the object in size and shape - to enable differentiation and accelerate the training process.
Applying different materials and textures, with random erosions, blur and noise to achieve the prevailing characteristics of a real image.
Applying variations of weather conditions and time of day based on requirements.
Adapting the dataset to a specific sensor by applying camera resolution, shooting angle and off-nadir angle. Can accommodate any photography system.
Customizing and optimizing the annotations to accommodate information added in previous layers.
No two cities are alike, nor fields or airports. Each is different and unique in multiple aspects.
To produce virtual images that perfectly mimic the randomness of the real world, we apply Structured Randomization to every aspect of the creation process. Every variable is treated with a dedicated randomization factor – color, texture, size, location, light – to produce truly random-real results.