Image processing method, image processing device and computer readable medium, for acquiring image sample data for training an attribute recognition model

Patent metadata

Field	Value
Publication number	US-11928183-B2
Application number	US-202117532537-A
Country	US
Kind code	B2
Filing date	Nov 22, 2021
Priority date	Jul 29, 2021
Publication date	Mar 12, 2024
Grant date	Mar 12, 2024

Abstract

An image processing method includes acquiring a set of image samples for training an attribute recognition model, wherein the set of image samples includes a first subset of image samples with category labels and a second subset of image samples without category labels; training a sample prediction model using the first subset of image samples, and predicting categories of the image samples in the second subset of image samples using the trained sample prediction model; determining a category distribution of the set of image samples based on the category labels of the first subset of image samples and the predicted categories of the second subset of image samples; and acquiring a new image sample if the determined category distribution does not conform to the expected category distribution, and adding the acquired new image sample to the set of image samples.

First claim

What is claimed is:

1. An image processing method, comprising:
acquiring a set of training image samples for training an attribute recognition model, wherein the set of training image samples comprises a first subset of training image samples with category labels and a second subset of training image samples without category labels;
training a sample prediction model using the first subset of training image samples with category labels, and predicting categories of the training image samples in the second subset of training image samples without category labels using the trained sample prediction model;
determining a category distribution of the set of training image samples based on the category labels of the first subset of training image samples and the predicted categories of the second subset of training image samples; and
acquiring a new training image sample directionally if the determined category distribution does not conform to a certain category distribution, and updating the set of training image samples by adding the acquired new training image sample to the set of training image samples so that the category distribution of the updated set of training image samples conforms to the certain category distribution, wherein the certain category distribution is a certain number or a certain ratio in the determined category distribution;
wherein the conforming of the determined category distribution to the certain category distribution comprises:
determining a category which does not reach a certain number or a certain ratio in the determined category distribution, which indicates the determined category distribution does not conform to the certain category distribution; and
acquiring the new training image sample of the category that does not reach the certain number or the certain ratio.

2. The image processing method of claim 1 , wherein the first subset of training image samples includes a predetermined number of training image samples for each category.

3. The image processing method according to claim 1 , wherein the first subset of training image samples is selected randomly from the set of training image samples, and the training image processing method further comprises acquiring the category labels of the first subset of training image samples.

4. The image processing method according to claim 1 , wherein the category labels of the first subset of training image samples are pre-assigned.

5. The image processing method of claim 4 , wherein data sources of the first subset of training image samples and the second subset of training image samples are different.

6. The image processing method according to claim 1 , wherein the sample prediction model is a Few-Shot Learning model.

7. The image processing method according to claim 6 , wherein the Few-Shot Learning model comprises at least one of a Pretraining model or a meta-learning model.

8. The image processing method according to claim 1 , wherein the certain category distribution includes at least one of Uniform distribution or Gaussian distribution.

9. The image processing method according to claim 1 , wherein the new training image sample is acquired by searching a database with a keyword or a picture.

10. The image processing method according to claim 1 , wherein the new training image sample is generated by picture editing.

11. The image processing method according to claim 1 , further comprising:
predicting a category of the new training image sample using the trained sample prediction model, and discarding the new training image sample if the predicted category of the new training image sample is not the determined category that does not reach the certain number or the certain ratio.

12. The image processing method according to claim 1 , further comprising:
redetermining the category distribution of the updated set of training image samples, and acquiring an additional new training image sample if the redetermined category distribution does not conform to the certain category distribution.

13. The image processing method according to claim 1 , further comprising:
acquiring category labels of the updated set of training image samples.

14. The image processing method according to claim 1 , further comprising:
training the attribute recognition model using the updated set of training image samples.

15. The image processing method according to claim 1 , wherein the category labels, the predicted categories and the category distribution belong to a particular attribute.

16. The image processing method according to claim 15 , wherein the particular attribute is an attribute of a human face.

17. An electronic device comprising:
a memory; and
a processor coupled to the memory, wherein instructions are stored in the memory, and the instructions when executed by the processor cause the electronic device to:
acquire a set of training image samples for training an attribute recognition model, wherein the set of training image samples comprises a first subset of training image samples with category labels and a second subset of training image samples without category labels;
train a sample prediction model using the first subset of training image samples with category labels, and predict categories of the training image samples in the second subset of training image samples without category labels using the trained sample prediction model;
determine a category distribution of the set of training image samples based on the category labels of the first subset of training image samples and the predicted categories of the second subset of training image samples; and
acquire a new training image sample directionally if the determined category distribution does not conform to a certain category distribution, and update the set of training image samples by adding the acquired new training image sample to the set of training image samples so that the category distribution of the updated set of training image samples conforms to the certain category distribution, wherein the certain category distribution is a certain number or a certain ratio in the determined category distribution;
wherein the conforming of the determined category distribution to the certain category distribution comprises:
determining a category which does not reach a certain number or a certain ratio in the determined category distribution which indicates the determined category distribution does not conform to the certain category distribution; and
acquiring the new training image sample of the category that does not reach the certain number or the certain ratio.

18. A non-transitory computer-readable storage medium with computer program stored thereon which, when executed by a processor, cause the processor to:
acquire a set of training image samples for training an attribute recognition model, wherein the set of training image samples comprises a first subset of training image samples with category labels and a second subset of training image samples without category labels;
train a sample prediction model using the first subset of training image samples with category labels, and predict categories of the training image samples in the second subset of training image samples without category labels using the trained sample prediction model;
determine a category distribution of the set of training image samples based on the category labels of the first subset of training image samples and the predicted categories of the second subset of training image samples; and
acquire a new training image sample directionally

Assignees

• Lemon Inc

Inventors

• Sun Jingna
• Zeng Weihong
• Chen Peibin
• Wang Xu
• Lai Chunpong
• Sang Shen
• Liu Jing

Classifications

• G06F18/2155 Primary

  characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling · CPC title

• G06F16/532

  Query formulation, e.g. graphical querying · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06V40/168

  Feature extraction; Face representation · CPC title

• G06V10/82 Primary

  using neural networks · CPC title