# US 2020/0349389 A1 · Method and device for training image recognition model and related device

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0349389 A1
- Document ID: US20200349389A1
- Status: Published application
- Kind: A1
- Application number: 16/828,845
- Filed: 2020-03-24
- Published: 2020-11-05
- Earliest claimed: 2019-04-30
- Assignees: Beijing Baidu Netcom Science Technology Co., Ltd.
- Applicants: Beijing Baidu Netcom Science and Technology Co., Ltd.
- Inventors: Jiadong ZHANG; Chao LI; Guoyi LIU
- Web page: https://patentgrep.com/patent/US20200349389A1

## CPC

- G06V 10/751: {Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06T 7/73: using feature-based methods
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06T 2207/20081: Training; Learning
- G06V 2201/07: Target detection
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

The present disclosure relates to a method and a device for training an image recognition model and a related device. The method includes: extracting sub-image feature data from a detection frame sub-image of an input image; determining element feature data matching the sub-image feature data from an index element database; and outputting images related to the element feature data as training images for training the image recognition model. The index element database is built in advance based on a plurality of element feature data extracted from a plurality of candidate images.

## Claims

### Claim 1

A method for training an image recognition model, comprising:
extracting sub-image feature data from a detection frame sub-image of an input image;
determining element feature data matching the sub-image feature data from an index element database, wherein the index element database is built in advance based on a plurality of element feature data extracted from a plurality of candidate images; and
outputting images related to the element feature data as training images for training the image recognition model.

### Claim 2 (depends on 1)

The method according to claim 1, wherein building the index element database comprises:
extracting the plurality of element feature data from respective detection frame sub-images of a plurality of labeled candidate images; and
storing the plurality of element feature data in the index element database.

### Claim 3 (depends on 1)

The method according to claim 1, wherein building the index element database comprises:
determining respective detection frame sub-images of a plurality of unlabeled candidate images;
extracting the plurality of element feature data from the respective detection frame sub-images of the plurality of unlabeled candidate images; and
storing the plurality of element feature data in the index element database.

### Claim 4 (depends on 2)

The method according to claim 2, wherein building the index element database further comprises:
generating mapping information amongst the element feature data, the respective candidate images and positions in the respective candidate images.

### Claim 5 (depends on 4)

The method according to claim 4, wherein outputting the images related to the element feature data as the training images, comprises:
for each element feature data, determining a candidate image and a position in the candidate image based on the mapping information; and
outputting the candidate image as the training image.

### Claim 6 (depends on 3)

The method according to claim 3, wherein building the index element database further comprises:
generating mapping information amongst the element feature data, the respective candidate images and positions in the respective candidate images.

### Claim 7 (depends on 6)

The method according to claim 6, wherein outputting images related to the element feature data as training images, comprises:
for each element feature data, determining a candidate image and a position in the candidate image based on the mapping information; and
outputting the candidate image as the training images.

### Claim 8 (depends on 1)

The method according to claim 1, wherein determining the element feature data matching the sub-image feature data from the index element database comprises:
comparing the sub-image feature data with the element feature data of the index element database; and
in response to that a similarity between the sub-image feature data and the element feature data is greater than a predetermined threshold, outputting the element feature data.

### Claim 9 (depends on 2)

The method according to claim 2, wherein the sub-image feature data and the plurality of element feature data are feature vectors, and determining the element feature data matching the sub-image feature data from the index element database comprises:
calculating a similarity between the sub-image feature data and each element feature data of the index element database; and
determining a predetermined number of the element feature data each having a high similarity as the element feature data matching the sub-image feature data.

### Claim 10 (depends on 3)

The method according to claim 3, wherein the sub-image feature data and the plurality of element feature data are feature vectors, and determining the element feature data matching the sub-image feature data from the index element database comprises:
calculating a similarity between the sub-image feature data and each element feature data of the index element database; and
determining a predetermined number of the element feature data each having a high similarity as the element feature data matching the sub-image feature data.

### Claim 11

An electronic device, comprising:
one or more processors; and
a storage device, configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to:
extract sub-image feature data from a detection frame sub-image of an input image;
determine element feature data matching the sub-image feature data from an index element database, wherein the index element database is built in advance based on a plurality of element feature data extracted from a plurality of candidate images; and
output images related to the element feature data as training images for training the image recognition model.

### Claim 12 (depends on 11)

The electronic device according to claim 11, wherein the one or more processors are caused to build the index element database by:
extracting the plurality of element feature data from respective detection frame sub-images of a plurality of labeled candidate images; and storing the plurality of element feature data in the index element database;
or
determining respective detection frame sub-images of a plurality of unlabeled candidate images; extracting the plurality of element feature data from the respective detection frame sub-images of the plurality of unlabeled candidate images; and storing the plurality of element feature data in the index element database.

### Claim 13 (depends on 12)

The electronic device according to claim 12, wherein the one or more processors are caused to build the index element database further by:
generating mapping information amongst the element feature data, the respective candidate images and positions in the respective candidate images.

### Claim 14 (depends on 13)

The electronic device according to claim 13, wherein the one or more processors are caused to output images related to the element feature data as the training images by:
for each element feature data, determining a candidate image and a position in the candidate image based on the mapping information; and
outputting the candidate image as the training image.

### Claim 15 (depends on 11)

The electronic device according to claim 11, wherein the one or more processors are configured to determine the element feature data matching the sub-image feature data from the index element database by:
comparing the sub-image feature data with the element feature data of the index element database; and
in response to that a similarity between the sub-image feature data and the element feature data is greater than a predetermined threshold, outputting the element feature data.

### Claim 16 (depends on 12)

The electronic device according to claim 12, wherein the sub-image feature data and the plurality of element feature data are feature vectors, and
the one or more processors are configured to determine the element feature data matching the sub-image feature data from the index element database by:
calculating a similarity between the sub-image feature data and each element feature data of the index element database; and
determining a predetermined number of the element feature data each having a high similarity as the element feature data matching the sub-image feature data.

### Claim 17

A non-transitory computer-readable storage medium, having a computer program stored thereon, wherein when the computer program is executed by a processor, a method for training an image recognition model is implemented, the method comprising:
extracting sub-image feature data from a detection frame sub-image of an input image;
determining element feature data matching the sub-image feature data from an index element database, wherein the index element database is built in advance based on a plurality of element feature data extracted from a plurality of candidate images; and
outputting images related to the element feature data as training images for training the image recognition model.

### Claim 18 (depends on 17)

The non-transitory computer-readable storage medium according to claim 17, wherein building the index element database comprises:
extracting the plurality of element feature data from respective detection frame sub-images of a plurality of labeled candidate images; and storing the plurality of element feature data in the index element database;
or
determining respective detection frame sub-images of a plurality of unlabeled candidate images; extracting the plurality of element feature data from the respective detection frame sub-images of the plurality of unlabeled candidate images; and storing the plurality of element feature data in the index element database.

### Claim 19 (depends on 18)

The non-transitory computer-readable storage medium according to claim 18, wherein building the index element database further comprises:
generating mapping information amongst the element feature data, the respective candidate images and positions in the respective candidate images.

### Claim 20 (depends on 19)

The non-transitory computer-readable storage medium according to claim 19, wherein outputting the images related to the element feature data as the training images, comprises:
for each element feature data, determining a candidate image and a position in the candidate image based on the mapping information; and
outputting the candidate image as the training image.
