| CPC G06V 40/172 (2022.01) [G06V 10/771 (2022.01); G06V 40/161 (2022.01); G06V 40/168 (2022.01); G06V 40/12 (2022.01)] | 14 Claims |
|
1. A method of performing object recognition, comprising:
by converting a first feature of a target object extracted using a first extraction model to a feature space of a second extraction model through a first feature conversion model, obtaining a second feature of the target object in the feature space;
by matching the second feature of the target object with features of objects in a matching library, obtaining a target feature matched with the second feature of the target object; and
determining an object to which the target feature belongs as a matching object of the target object;
wherein the method further comprises:
obtaining features of a plurality of sample objects extracted through the first extraction model and through the second extraction model;
by inputting features of a target number of first sample objects in the plurality of sample objects extracted through the first extraction model into a first initial feature conversion model, obtaining a first output result;
determining a first loss value between the first output result and features of the target number of first sample objects extracted through the second extraction model; and
according to the first loss value, the first initial feature conversion model, the features of the plurality of sample objects extracted through the first extraction model and through the second extraction model, determining the first feature conversion model;
wherein according to the first loss value, the first initial feature conversion model, the features of the plurality of sample objects extracted through the first extraction model and through the second extraction model, determining the first feature conversion model, comprising:
by taking the first loss value as a constraint, adjusting parameter values of parameters in the first initial feature conversion model based on a gradient descent algorithm;
re-selecting a target number of second sample objects from the plurality of sample objects, obtaining another output result by inputting features of the target number of second sample objects extracted through the first extraction model into the first initial feature conversion model, determining another loss value between the another output result and features of the target number of second sample objects extracted through the second extraction model;
continuing to adjust the parameter values of the parameters in the first initial feature conversion model until a minimum loss value is obtained, and obtaining the first feature conversion model by substituting parameter values of parameters corresponding to the minimum loss value into the first initial feature conversion model.
|