CPC G06F 18/214 (2023.01) [G06F 18/22 (2023.01); G06T 7/70 (2017.01); G06T 17/00 (2013.01); G06T 19/20 (2013.01); G06T 2207/20081 (2013.01); G06T 2219/004 (2013.01); G06T 2219/2008 (2013.01); G06T 2219/2012 (2013.01)] | 18 Claims |
1. A method performed by data processing apparatus, the method comprising:
receiving a plurality of real images of one or more instances of an object, wherein, in each of the real images, the imaged instance has a respective pose;
receiving a three-dimensional model of the object, wherein locations of landmarks of the object are derivable from the three-dimensional model;
creating a plurality of surrogate images of instances of the object using the three-dimensional model of the object, wherein creating each of the surrogate images comprises
rendering the three-dimensional model of the object in two dimensions in one of the poses,
perturbing a characteristic of the rendering of the model of the object in two dimensions, and
labeling landmarks on the rendering of the model of the object in two dimensions based on the locations derived from the three-dimensional model; and
training or retraining a machine learning pose estimation model using the collection of surrogate images of the object.
|