Granted Patent B2

US 12,482,225 · App. 17/990,366 · Granted Nov 25, 2025

Inventors: Zijia Wang (WeiFang, CN); Jinpeng Liu (Shanghai, CN); Jiacheng Ni (Shanghai, CN); Zhen Jia (Shanghai, CN)

Assignee: Dell Products L.P.

G06V10/761G06V10/764G06V10/7715G06V10/774G06V10/7796G06V10/82

Abstract

Claims (79)

1 . A method for acquiring an image, comprising:

distilling an original image set through a capsule neural network model to generate a distilled image set, the distilled image set comprising a plurality of distilled images;

acquiring a first feature of a first image through the capsule neural network model;

acquiring a plurality of distilling features of the plurality of distilled images respectively through the capsule neural network model;

determining a plurality of similarities between the first feature and the plurality of distilling features respectively; and

acquiring at least one original image matching the first image based on the plurality of similarities;

wherein generating the distilled image set through the capsule neural network model comprises:

inputting each original image in the original image set to the capsule neural network model and acquiring a set of output features;

acquiring a set of capsule features of the capsule neural network model; and

acquiring the distilled image set based on the set of output features and the set of capsule features;

wherein the set of output features comprises an output feature matrix, and wherein an element in a column vector of the output feature matrix indicates a probability that each original image in the original image set belongs to a corresponding target class.

2 . The method according to claim 1 , wherein acquiring the at least one original image matching the first image comprises:

determining a maximum similarity among the plurality of similarities;

using a distilled image corresponding to the maximum similarity as a reference distilled image;

determining an original image in the original image set having a maximum weight relative to the reference distilled image; and

using the original image having the maximum weight as the original image matching the first image.

3 . The method according to claim 1 , wherein acquiring the at least one original image matching the first image comprises:

determining a soft label of the first image based on the plurality of similarities, wherein the soft label comprises a plurality of elements, and each element indicates a probability that the first image belongs to each of a plurality of target classes; and

acquiring the at least one original image based on the soft label.

4 . The method according to claim 1 , wherein generating the distilled image set through the capsule neural network model further comprises:

acquiring a column vector in the output feature matrix that corresponds to the corresponding target class;

determining a probability value corresponding to each distilled image based on elements in the column vector; and

determining, based on the probability value corresponding to each distilled image, a probability that each distilled image is under the corresponding target class as a soft label of the distilled image.

5 . The method according to claim 4 , wherein determining, based on the probability value corresponding to each distilled image, the probability that each distilled image is under the corresponding target class comprises:

determining the number of original images corresponding to the distilled image corresponding to each probability value; and

determining, based on each probability value and the number, the probability that each distilled image is under the corresponding target class.

6 . The method according to claim 1 , further comprising:

receiving a second image and a third image; and

determining a similarity between the second image and the third image through the capsule neural network model, wherein the distilled image set does not include the second image and the third image.

7 . The method according to claim 1 , wherein the first feature and the plurality of distilling features are extracted through a feature extracting module in the capsule neural network model, and wherein an input to and an output from the feature extracting module are in residual connection.

8 . The method according to claim 1 , further comprising:

training the capsule neural network model based on a triplet loss function.

9 . An electronic device, comprising:

at least one processor; and

memory coupled to the at least one processor and having instructions stored thereon, wherein the instructions, when executed by the at least one processor, cause the electronic device to perform actions comprising:

distilling an original image set through a capsule neural network model to generate a distilled image set, the distilled image set comprising a plurality of distilled images;

acquiring a first feature of a first image through the capsule neural network model;

acquiring a plurality of distilling features of the plurality of distilled images respectively through the capsule neural network model;

determining a plurality of similarities between the first feature and the plurality of distilling features respectively; and

acquiring at least one original image matching the first image based on the plurality of similarities;

wherein generating the distilled image set through the capsule neural network model comprises:

inputting each original image in the original image set to the capsule neural network model and acquiring a set of output features;

acquiring a set of capsule features of the capsule neural network model; and

acquiring the distilled image set based on the set of output features and the set of capsule features;

wherein the set of output features comprises an output feature matrix, and wherein an element in a column vector of the output feature matrix indicates a probability that each original image in the original image set belongs to a corresponding target class.

10 . The electronic device according to claim 9 , wherein acquiring the at least one original image matching the first image comprises:

determining a maximum similarity among the plurality of similarities;

using a distilled image corresponding to the maximum similarity as a reference distilled image;

determining an original image in the original image set having a maximum weight relative to the reference distilled image; and

using the original image having the maximum weight as the original image matching the first image.

11 . The electronic device according to claim 9 , wherein acquiring the at least one original image matching the first image comprises:

determining a soft label of the first image based on the plurality of similarities, wherein the soft label comprises a plurality of elements, and each element indicates a probability that the first image belongs to each of a plurality of target classes; and

acquiring the at least one original image based on the soft label.

12 . The electronic device according to claim 9 , wherein generating the distilled image set through the capsule neural network model further comprises:

acquiring a column vector in the output feature matrix that corresponds to the corresponding target class;

determining a probability value corresponding to each distilled image based on elements in the column vector; and

determining, based on the probability value corresponding to each distilled image, a probability that each distilled image is under the corresponding target class as a soft label of the distilled image.

13 . The electronic device according to claim 12 , wherein determining, based on the probability value corresponding to each distilled image, the probability that each distilled image is under the corresponding target class comprises:

determining the number of original images corresponding to the distilled image corresponding to each probability value; and

determining, based on each probability value and the number, the probability that each distilled image is under the corresponding target class.

14 . The electronic device according to claim 9 , wherein the instructions, when executed by the at least one processor, cause the electronic device to perform actions comprising:

receiving a second image and a third image; and

determining a similarity between the second image and the third image through the capsule neural network model, wherein the distilled image set does not include the second image and the third image.

15 . The electronic device according to claim 9 , wherein the first feature and the plurality of distilling features are extracted through a feature extracting module in the capsule neural network model, and wherein an input to and an output from the feature extracting module are in residual connection.

16 . A computer program product that is tangibly stored on a non-transitory computer-readable medium and comprises machine-executable instructions, wherein the machine-executable instructions, when executed by a machine, cause the machine to perform actions comprising:

distilling an original image set through a capsule neural network model to generate a distilled image set, the distilled image set comprising a plurality of distilled images;

acquiring a first feature of a first image through the capsule neural network model;

acquiring a plurality of distilling features of the plurality of distilled images respectively through the capsule neural network model;

determining a plurality of similarities between the first feature and the plurality of distilling features respectively; and

acquiring at least one original image matching the first image based on the plurality of similarities;

wherein generating the distilled image set through the capsule neural network model comprises:

inputting each original image in the original image set to the capsule neural network model and acquiring a set of output features;

acquiring a set of capsule features of the capsule neural network model; and

acquiring the distilled image set based on the set of output features and the set of capsule features;

wherein the set of output features comprises an output feature matrix, and wherein an element in a column vector of the output feature matrix indicates a probability that each original image in the original image set belongs to a corresponding target class.

17 . The method according to claim 1 , wherein the capsule neural network model comprises a feature extraction portion coupled to a capsule layer portion, the capsule layer portion comprising a main capsule layer and a classifying capsule layer.

18 . The electronic device according to claim 9 , wherein the capsule neural network model comprises a feature extraction portion coupled to a capsule layer portion, the capsule layer portion comprising a main capsule layer and a classifying capsule layer.

19 . The method according to claim 1 , wherein the output feature matrix has a number of rows that corresponds to a number n of original images in the original image set, and a number of columns that corresponds to a number k of capsules in the capsule neural network model.

20 . The electronic device according to claim 9 , wherein the output feature matrix has a number of rows that corresponds to a number n of original images in the original image set, and a number of columns that corresponds to a number k of capsules in the capsule neural network model.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (10)