Apparatus and method for re-identifying object
AssigneesHyundai Motor Company, Kia Corporation, Industry-Academic Cooperation Foundation, Yonsei University
InventorsKyung Hwan Cho, Moon Sub Jin, Hyeon Seok Jung, Bum Sub Ham, Chan Ho Eom, Geon Lee
Life of this patent
Estimating the term…
Abstract
An apparatus and a method for re-identifying an object to improve object re-identification performance using attribute information of the object are provided. The apparatus trains an object representation extraction model to train attribute representations, inputs an image obtained by a camera to the trained object representation extraction model, extracts object representations from the image using the trained object representation extraction model, and performs object re-identification based on the object representations.
Claims16
An apparatus comprising:
- a processor;
- a non-transitory storage medium coupled to the processor, the storage medium storing instructions that, when executed by the processor, cause the processor to:
- train an object representation extraction model to learn attribute representations;
- input an image obtained by a camera to the trained object representation extraction model;
- extract object representations from the image using the trained object representation extraction model;
- and
- perform object re-identification based on the object representations,
- wherein the storage medium stores further instructions that, when executed by the processor, further cause the processor to:
- train the object representation extraction model using a loss function so that the object representations are trained to be similar to an attribute prototype corresponding to the object representations,
- wherein the loss function considers an object ID label of an ith image, ith image object representations, and a probability of accurately predicting the object ID label of the ith image using the ith image object representations,
- wherein the storage medium stores instructions that, when executed by the processor, further cause the processor to:
- limit a similarity between the object representations and the attribute representations within an embedding space such that the similarity is not further increased once the similarity exceeds a predetermined threshold in the loss function, to prevent a degradation in re-identification performance caused by collapse or over-convergence of representation vectors of different objects with identical attributes.
A method comprising:
- training, by a processor, an object representation extraction model to train attribute representations;
- inputting, by the processor, an image obtained by a camera to the trained object representation extraction model;
- extracting, by the processor, object representations from the image using the trained object representation extraction model;
- and
- performing, by the processor, object re-identification based on the object representations,
- training, by the processor, the object representation extraction model using a loss function so that the object representations are trained to be similar to an attribute prototype corresponding to the object representations,
- wherein the loss function considers an object ID label of an ith image, ith image object representations, and a probability of accurately predicting the object ID label of the ith image using the ith image object representations;
- determining that a similarity between the object representations and the attribute representations within an embedding space exceeds a predetermined threshold in the loss function;
- and
- limiting, by the processor, the similarity between the object representations and the attribute representations within the embedding space, such that the similarity is not further increased, in response to determining that the similarity exceeds the predetermined threshold in the loss function, to prevent a degradation in re-identification performance caused by collapse or over-convergence of representation vectors of different objects with identical attributes.
Dependencies are inferred from the claim text.