Method for training an object recognition model in a computing device
An object recognition model training method in a computing device is disclosed. In the present disclosure, an object of interest, which is an object for object recognition, is designated, and an object of non-interest excluding the object of interest is generated and used as learning data for the object recognition model. In the process of training the object recognition model, when an erroneously detected object occurs, the object recognition model may be retrained by automatically converting the erroneously detected object to the object of non-interest without feedback of the erroneous detection to the user. Accordingly, user convenience for processing the erroneously detected object is improved, which increases reliability of the object recognition model. This disclosure can be associated with artificial intelligence modules, drones (unmanned aerial vehicles (UAVs)), robots, augmented reality (AR) devices, virtual reality (VR) devices, devices related to 5G service, etc.
1 . An object recognition model training method in a computing device, the method comprising:
receiving an input for designating an object of interest in an image acquired through a camera, wherein the input for designating the object of interest is received as a user selects displayed objects in the image;
generating an object of non-interest by designating at least a portion of a region excluding the object of interest in the image;
training an object recognition model based on the object of interest and the object of non-interest;
when a first object not designated as the object of interest is recognized as a result of performing object recognition based on the trained object recognition model, changing the first object to the object of non-interest; and
retraining the object recognition model based on the changed object of non-interest, wherein the generating of the object of non-interest includes generating N number of sets of objects of non-interest with different attributes, and
wherein the N number of sets of objects of non-interest includes a first set of objects of non-interest randomly designated in the region excluding the object of interest by a processor of the computing device, and a plurality of second sets of objects of non-interest generated while, among grid regions obtained by dividing the image at a predetermined grid interval, changing the grid interval in the region excluding the object of interest.
2 . The object recognition model training method of claim 1 , wherein the receiving of the input for designating the object of interest includes extracting location information of the object of interest as the object of interest is designated.
3 . The object recognition model training method of claim 1 , wherein
the plurality of second sets of objects of non-interest include a plurality of pixel regions designated in a state where the grid interval is adjusted to a first grid interval and a plurality of pixel regions designated in a state where the grid interval is adjusted to a second grid interval.
4 . The object recognition model training method of claim 1 , wherein in the training of the object recognition model, each of N number of object recognition models is trained using the object of interest and each of the N number of sets of objects of non-interest as a pair of learning data.
5 . The object recognition model training method of claim 1 , wherein in the generating of the object of non-interest, the object of non-interest is additionally designated when a confidence score in the region excluding the object of interest is similar to the object of interest within a predetermined range, based on the confidence score of a previous training model in a process of repeatedly performing the training of the object recognition model.
6 . The object recognition model training method of claim 1 , wherein the changing of the first object to the object of non-interest includes: acquiring object characteristic information of the first object; and selecting a second object to be changed to the object of non-interest based on the object characteristic information of the first object among objects not generated as the object of non-interest, and additionally changing the second object to the object of non-interest.
7 . The object recognition model training method of claim 1 , further comprising: changing and displaying a visual characteristic of a bounding box of the object changed to the object of non-interest.
8 . A computing device comprising:
a communication unit that receives an image acquired through a camera;
a user input unit that designates an object of interest in the image;
a processor that generates an object of non-interest by designating at least a portion of a region excluding the object of interest in the image and trains an object recognition model based on the object of interest and the object of non-interest;
wherein, in case that a first object not designated as the object of interest is recognized as a result of performing object recognition based on the trained object recognition model, the processor changes the first object to the object of non-interest and retrains the object recognition model based on the changed object of non-interest, wherein the processor generates N number of sets of objects of non-interest with different attributes, and
wherein the N number of sets of objects of non-interest includes a first set of objects of non-interest set randomly designated in the region excluding the object of interest by the processor of the computing device, and
a plurality of second sets of objects of non-interest sets generated while, among grid regions obtained by dividing the image at a predetermined grid interval, changing the grid interval in the region excluding the object of interest.
9 . The computing device of claim 8 , wherein the processor extracts location information of a bounding box of the object of interest as the object of interest is designated.
10 . The computing device of claim 8 , wherein The processor generates a plurality of pixel regions designated in a state in which the grid interval is adjusted to a first grid interval, and a plurality of pixel regions designated in a state in which the grid interval is adjusted to a second grid interval as the second sets of objects of non-interest, respectively.
11 . The computing device of claim 8 , wherein the processor additionally designates the object of non-interest when a confidence score in the region excluding the object of interest is similar to the object of interest within a predetermined range based on the confidence score of a previous learning model in a process of repeatedly training the object recognition model.
12 . The computing device of claim 8 , wherein the processor acquires object characteristic information of the first object, selects a second object to be changed to the object of non-interest based on the object characteristic information of the first object among objects not generated as the object of non-interest, and additionally changes the second object to the object of non-interest.
13 . A computing device comprising:
a communication unit that receives an image acquired through a camera;
a user input unit that receives a user selection of displayed objects in the image and designates an object of interest in the image in accordance with the user selection;
a processor that generates an object of non-interest by designating randomly selected portion of a region excluding the object of interest in the image and trains an object recognition model based on the object of interest and the object of non-interest;
wherein, in case that a first object not designated as the object of interest is recognized as a result of performing object recognition based on the trained object recognition model, the processor changes the first object to the object of non-interest and retrains the object recognition model based on the changed object of non-interest.