Neural network learning method and device for recognizing class

1. A neural network training method for recognizing a class, the neural network training method comprising:

acquiring, by using a first learning network model trained based on source training images respectively included in at least one class, feature information of a query image included in a class different from the at least one class;

obtaining a generated image from the feature information of the query image by using a second learning network model;

acquiring feature information of the obtained generated image by using the first learning network model; and

updating weights of layers respectively included in the first and second learning network models, based on a difference between the feature information of the query image and the feature information of the generated image and on a difference between the query image and the generated image.

2. The neural network training method of claim 1 , further comprising acquiring information about a target class,

wherein the obtaining of the generated image comprises obtaining, based on the feature information of the query image and the information about the target class, a generated image included in the target class by using the second learning network model.

3. The neural network training method of claim 1 , wherein the updating comprises iteratively performing the updating until the difference between the feature information of the query image and the feature information of the generated image and the difference between the query image and the generated image are acquired below thresholds respectively preset with respect to the differences therebetween.

4. The neural network training method of claim 3 , further comprising determining, as a training image for a recognition model for class recognition, a generated image obtained via a second learning network model acquired as a result of the iterative performing of the updating.

5. The neural network training method of claim 1 , wherein the acquiring of the feature information of the generated image comprises acquiring feature information of a generated image having a smallest difference from the query image from among a plurality of generated images obtained based on the feature information of the query image.

6. The neural network training method of claim 1 , wherein the acquiring of the feature information of the generated image comprises receiving, from a server, the feature information of the query image acquired by using the first learning network model stored in the server,

the neural network training method further comprising transmitting, to the server, the generated image obtained using the second learning network model.

7. A neural network training device for recognizing a class, the neural network training device comprising:

a memory storing one or more instructions; and

at least one processor configured to execute the one or more instructions stored in the memory to:

acquire, by using a first learning network model trained based on source training images respectively included in at least one class, feature information of a query image included in a class different from the at least one class;

obtain a generated image from the feature information of the query image by using a second learning network model;

acquire feature information of the obtained generated image by using the first learning network model; and

update weights of layers respectively included in the first and second learning network models, based on a difference between the feature information of the query image and the feature information of the generated image and on a difference between the query image and the generated image.

8. The neural network training device of claim 7 , wherein the at least one processor is further configured to execute the one or more instructions to:

acquire information about a target class; and

obtain, based on the feature information of the query image and the information about the target class, a generated image included in the target class by using the second learning network model.

9. The neural network training device of claim 7 , wherein the at least one processor is further configured to execute the one or more instructions to iteratively perform the updating until the difference between the feature information of the query image and the feature information of the generated image and the difference between the query image and the generated image are acquired below thresholds respectively preset with respect to the differences therebetween.

10. The neural network training device of claim 9 , wherein the at least one processor is further configured to execute the one or more instructions to determine, as a training image for a recognition model for class recognition, a generated image obtained via a second learning network model acquired as a result of the iterative performing of the updating.

11. The neural network training device of claim 7 , wherein the at least one processor is further configured to execute the one or more instructions to acquire feature information of a generated image having a smallest difference from the query image from among a plurality of generated images obtained based on the feature information of the query image.

12. The neural network training device of claim 7 , further comprising a communication unit,

wherein the at least one processor is further configured to execute the one or more instructions to control the communication unit to receive, from a server, the feature information of the query image acquired by using the first learning network model stored in the server and to transmit, to the server, the generated image obtained using the second learning network model stored in the neural network training device.

13. A computer-readable recording medium having recorded thereon a program for executing the neural network training method of claim 1 on a computer.