US 12,147,496 B1
Automatic generation of training data for instance segmentation algorithms
Francesco Giuseppe Callari, Seattle, WA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Nov. 3, 2021, as Appl. No. 17/518,012.
Int. Cl. G06T 7/70 (2017.01); G06F 18/214 (2023.01); G06T 19/20 (2011.01)
CPC G06F 18/2148 (2023.01) [G06T 7/70 (2017.01); G06T 19/20 (2013.01); G06T 2219/2004 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A method to automatically generate training data, comprising:
receiving, by a controller from an imaging device associated with an aerial vehicle, imaging data;
receiving, by the controller, position and orientation data associated with the imaging device;
receiving, by the controller, position and orientation data associated with an airborne object represented within the imaging data;
receiving, by the controller, attribute data associated with the airborne object;
receiving, by the controller, a three-dimensional model corresponding to the airborne object based on the attribute data;
rendering, by the controller, the three-dimensional model overlaying the airborne object represented within the imaging data based on the position and orientation data of the imaging device and the position and orientation data of the airborne object;
generating, by the controller, a mask based on the three-dimensional model overlaying the airborne object; and
providing, by the controller, the imaging data and the mask to a machine learning model as training data.