US 12,136,293 B2
Image group classifier in a user device
Shaomin Xiong, Newark, CA (US); and Toshiki Hirano, San Jose, CA (US)
Assigned to Sandisk Technologies, Inc., Milpitas, CA (US)
Filed by Western Digital Technologies, Inc., San Jose, CA (US)
Filed on Dec. 21, 2021, as Appl. No. 17/557,541.
Prior Publication US 2023/0196831 A1, Jun. 22, 2023
Int. Cl. G06V 40/16 (2022.01); G06V 10/70 (2022.01); G06V 20/52 (2022.01)
CPC G06V 40/172 (2022.01) [G06V 10/70 (2022.01); G06V 20/53 (2022.01); G06V 40/168 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A device comprising:
at least one processor;
at least one memory;
a storage medium configured to store user image data;
an image manager stored in the at least one memory for execution by the at least one processor alone or in combination, wherein the image manager is configured to index image data objects in the user image data;
a calibration engine stored in the at least one memory for execution by the at least one processor alone or in combination, wherein the calibration engine is configured to:
detect an object type in a plurality of image data objects in the user image data;
extract, for each image data object including the object type, embeddings for that image data object;
determine, using a clustering algorithm and the embeddings for image data objects including the object type, a plurality of initial groups of image data objects with clustered embeddings;
determine a set of group selection criteria;
select, using the set of group selection criteria, a first group from the plurality of initial groups;
determine a first group classifier;
assign a first group identifier for the first group to the first group classifier; and
train, using image data corresponding to the first group, the first group classifier to determine a set of machine learning parameters for the first group classifier to assign the first group identifier; and
at least one image group classifier stored in the at least one memory for execution by the at least one processor alone or in combination, wherein the at least one image group classifier is configured to:
determine embeddings quantifying a plurality of features in a new image data object;
process the embeddings for the new image data object through the set of machine learning parameters to assign the first group identifier; and
determine a confidence score for the new image data object corresponding to the first group identifier, wherein the image manager is further configured to selectively associate the first group identifier with the new image data object.