Device and method for determining convolutional neural network model for database

1. A device for determining a convolutional neural network (CNN) model for a database, comprising:

a processor configured to

select at least two CNN models from multiple CNN models whose classification capacity is known,

train the at least two CNN models to obtain first parameters of the at least two CNN models, respectively,

fit, based on the classification capacity and the first parameters of the at least two CNN models, a curve taking the classification capacity and the first parameters as variables, the first parameters characterizing performance of the at least two CNN models, respectively,

predict, based on the curve, a first parameter of each of the multiple CNN models other than the at least two CNN models, and

determine an applicable CNN model applicable to the database from the multiple CNN models based on the first parameters of the multiple CNN models.

2. The device according to claim 1 , wherein the processor is further configured to

determine, from the multiple CNN models, a first model among the multiple CNN models having a first classification capacity that best matches the database,

determine a second model from among the multiple CNN models, having a second classification capacity that is poorest,

determine the at least two CNN models, where the at least two CNN models include the first model and the second model.

3. The device according to claim 2 , wherein the processor is further configured to determine, from the multiple CNN models, a third model having a third classification capacity that is best, and the at least two CNN models determined by the processor further include the third model.

4. The device according to claim 1 , wherein the processor is further configured to fit the curve by one of a power exponent function and a logarithmic function.

5. The device according to claim 1 , wherein the first parameter includes a precision parameter for characterizing classification precision of the CNN model.

6. The device according to claim 1 , wherein the processor is further configured to determine, from the multiple CNN models, the applicable CNN model applicable to the database based on the first parameter and a second parameter of each of the multiple CNN models, the second parameter being used for characterizing performance of each of the multiple CNN models.

7. The device according to claim 6 , wherein the second parameter includes one of a speed parameter for characterizing a calculation speed of each of the multiple CNN models and an occupancy rate parameter for characterizing a memory occupancy rate of each of the multiple CNN models.

8. The device according to claim 7 , wherein the speed parameter for characterizing the calculation speed of each of the multiple CNN models includes a number of floating-point calculations of each of the multiple CNN models.

9. A method for determining a convolutional neural network (CNN) model for a database, comprising:

selecting at least two CNN models from multiple CNN models whose classification capacity is known;

training each of the at least two CNN models, to obtain the first parameters of the at least two CNN models, respectively;

fitting, based on the classification capacity and the first parameters of the at least two CNN models, a curve taking the classification capacity and the first parameters as variables, the first parameters characterizing performance of the CNN model;

predicting, based on the curve, a first parameter of each of the multiple CNN models other than the at least two CNN models; and

determining an applicable CNN model applicable to the database from the multiple CNN models based on the first parameter of each of the multiple CNN models.

10. The method according to claim 9 , wherein the selecting the at least two CNN models includes:

determining, from the multiple CNN models, a first model having a first classification capacity that best matches the database;

determining, from the multiple CNN models, a second model having a second classification capacity that is poorest; and

determining the at least two CNN models, where the at least two CNN models include the first model and the second model.

11. The method according to claim 10 , wherein the selecting the at least two CNN models further includes:

determining, from the multiple CNN models, one of the models whose classification capacity is a best as a third model, and the at least two CNN models further include the third model.

12. The method according to claim 9 , wherein the curve is fitted by one of a power exponent function and a logarithmic function.

13. The method according to claim 9 , wherein the first parameter includes a precision parameter for characterizing classification precision of the CNN model.

14. The method according to claim 9 , wherein the determining the CNN model applicable to the database includes: determining, from the multiple CNN models, the applicable CNN model applicable to the database based on the first parameter and a second parameter of each of the multiple CNN models, the second parameter being used for characterizing performance of each of the multiple CNN models.

15. The method according to claim 14 , wherein the second parameter includes one of a speed parameter for characterizing a calculation speed of each of the multiple CNN models and an occupancy rate parameter for characterizing a memory occupancy rate of each of the multiple CNN models.

16. The method according to claim 15 , wherein the speed parameter for characterizing the calculation speed of each of the multiple CNN models includes a number of floating-point calculations of each of the multiple CNN models.

17. A non-transitory machine-readable storage medium in which a program product including machine-readable instruction codes stored therein is carried, wherein when being read and executed by a computer, the machine-readable instruction codes enables the computer to perform the method according to claim 9 .