CPC G06N 3/04 (2013.01) [G06F 18/22 (2023.01); G06F 18/2415 (2023.01); G06N 3/08 (2013.01); G06N 20/10 (2019.01); G06V 10/454 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06V 20/40 (2022.01)] | 31 Claims |
1. A method of training and using a neural network, the method comprising:
generating a plurality of triplets of training vectors, each triplet in the plurality of triplets comprising a reference data point, a positive data point, and a negative data point, the reference data point representing a first object, the positive data point representing a second object similar to the first object, and the negative data point representing a third object dissimilar to the first object;
for each triplet in the plurality of triplets:
passing the reference data point, the positive data point, and the negative data point in each triplet through the neural network to generate extracted features;
calculating a loss from the extracted features; and
adjusting parameters of the neural network based on the loss;
generating, by the neural network, a feature embedding for an object appearing in an image shown on a display;
performing a comparison of the feature embedding to feature embeddings of other objects generated by the neural network; and
returning, based on the comparison, a representation of an object similar to the object appearing in the image shown in the display.
|