CPC G06T 7/75 (2017.01) [G06F 18/214 (2023.01); G06V 10/40 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] | 42 Claims |
1. A computer-implemented method for computing a pose of a deformable object, the method comprising:
receiving a plurality of images depicting the deformable object from multiple viewpoints;
computing a class of the deformable object depicted in the images;
loading a 3-D model corresponding to the class of the deformable object;
aligning the 3-D model to the deformable object depicted in the plurality of images to compute a pose of the deformable object, including:
refining a plurality of vertex positions of a plurality of vertices of the 3-D model to match appearances of the deformable object in the images captured from the multiple viewpoints,
wherein refining the plurality of vertex positions is performed using a loss function E comprising a first component E1 and a second component E2,
wherein E1 represents pixel-level differences between an appearance of the deformable object in the images and an appearance of the deformable object in renderings of the 3-D model from viewpoints corresponding to the images, and
wherein E2 represents a degree of alignment of a silhouette of the deformable object in the images and a silhouette of the deformable object in the renderings of the 3-D model from the viewpoints corresponding to the images; and
outputting the pose of the deformable object.
|