Image generation method, neural network compression method, and related apparatus and device
The present application discloses an image generation method, a neural network compression method, and a related apparatus and device in the field of artificial intelligence. The image generation method includes: inputting a first matrix into an initial image generator to obtain a generated image; inputting the generated image into a preset discriminator to obtain a determining result, where the preset discriminator is obtained through training based on a real image and a category corresponding to the real image; updating the initial image generator based on the determining result to obtain a target image generator; and further inputting a second matrix into the target image generator to obtain a sample image. Further, a neural network compression method is disclosed, to compress the preset discriminator based on the sample image obtained by using the foregoing image generation method.
1. An image generation method, comprising:
inputting a first matrix into an initial image generator to obtain a generated image, wherein the initial image generator is a deep neural network, and the first matrix does not contain image data;
inputting the generated image into a preset discriminator to obtain a determining result, wherein the preset discriminator is obtained through training based on first training data, and the first training data comprises a real image and a category corresponding to the real image;
updating the initial image generator based on the determining result to obtain a target image generator; and
inputting a second matrix into the target image generator to obtain a sample image, wherein the second matrix is a same type and format as the first matrix.
2. The method according to claim 1 , wherein the determining result comprises a probability that the generated image is predicted as each of M categories, and M is an integer greater than 1.
3. The method according to claim 2 , wherein the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; and
updating the initial image generator based on a difference between the determining result and the real result.
4. The method according to claim 2 , wherein before the updating the initial image generator based on the determining result, the method further comprises:
extracting a feature of the generated image by using the preset discriminator; and
the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; and updating the initial image generator based on the feature and a difference between the determining result and the real result.
5. The method according to claim 2 , wherein the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability;
obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results, wherein N is a positive integer; and
updating the initial image generator based on the average probability value and a difference between the determining result and the real result.
6. The method according to claim 2 , wherein before the updating the initial image generator based on the determining result, the method further comprises:
extracting a feature of the generated image by using the preset discriminator; and
the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results; and updating the initial image generator based on a difference between the determining result and the real result, an eigenvalue of the feature, and the average probability value.
7. A sample generation method, comprising:
inputting a first matrix into an initial sample generator to obtain a first generated sample, wherein the initial sample generator is a deep neural network, and the first matrix does not contain image data;
inputting the first generated sample into a preset discriminator to obtain a determining result, wherein the preset discriminator is obtained through training based on first training data, and the first training data comprises a real sample and a category corresponding to the real sample;
updating parameters of the initial sample generator based on the determining result of the first generated sample to obtain a target sample generator; and
inputting a second matrix into the target sample generator to obtain a second generated sample, wherein the second matrix is a same type and format as the first matrix.
8. The method according to claim 7 , wherein the determining result comprises a probability that the first generated sample is predicted as each of M categories, and M is an integer greater than 1.
9. The method according to claim 8 , wherein the updating the initial sample generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the first generated sample, a category corresponding to the maximum probability; and
updating the initial sample generator based on a difference between the determining result and the real result.
10. The method according to claim 8 , wherein before the updating the initial sample generator based on the determining result, the method further comprises:
extracting a feature of the first generated sample by using the preset discriminator; and
the updating the initial sample generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the first generated sample, a category corresponding to the maximum probability; and updating the initial sample generator based on the feature and a difference between the determining result and the real result.
11. The method according to claim 8 , wherein the updating the initial sample generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the first generated sample, a category corresponding to the maximum probability;
obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results, wherein N is a positive integer; and
updating the initial sample generator based on the average probability value and a difference between the determining result and the real result.
12. The method according to claim 8 , wherein before the updating the initial sample generator based on the determining result, the method further comprises:
extracting a feature of the first generated sample by using the preset discriminator; and
the updating the initial sample generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the first generated sample, a category corresponding to the maximum probability; obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results; and updating the initial sample generator based on a difference between the determining result and the real result, the feature, and the average probability value.
13. An image generation apparatus, comprising a memory and a processor, wherein the memory is configured to store a program, the processor is configured to execute the program stored in the memory, and when the program stored in the memory is executed, the processor is configured to perform an image generation method, the method comprising:
inputting a first matrix into an initial image generator to obtain a generated image, wherein the initial image generator is a deep neural network, and the first matrix does not contain image data;
inputting the generated image into a preset discriminator to obtain a determining result, wherein the preset discriminator is obtained through training based on first training data, and the first training data comprises a real image and a category corresponding to the real image;
updating the initial image generator based on the determining result to obtain a target image generator; and
inputting a second matrix into the target image generator to obtain a sample image, wherein the second matrix is a same type and format as the first matrix.
14. The apparatus according to claim 13 , wherein the determining result comprises a probability that the generated image is predicted as each of M categories, and M is an integer greater than 1.
15. The apparatus according to claim 14 , wherein the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; and
updating the initial image generator based on a difference between the determining result and the real result.
16. The apparatus according to claim 14 , wherein before the updating the initial image generator based on the determining result, the method further comprises:
extracting a feature of the generated image by using the preset discriminator; and
the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; and updating the initial image generator based on the feature and a difference between the determining result and the real result.
17. The apparatus according to claim 14 , wherein the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability;
obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results, wherein N is a positive integer; and
updating the initial image generator based on the average probability value and a difference between the determining result and the real result.
18. The apparatus according to claim 14 , wherein before the updating the initial image generator based on the determining result, the method further comprises:
extracting a feature of the generated image by using the preset discriminator; and
the updating the initial image generator based on the determining result specifically comprises:
determining a maximum probability in probabilities corresponding to the M categories, and determining, as a real result of the generated image, a category corresponding to the maximum probability; obtaining, based on N determining results that one-to-one correspond to N first matrices, an average probability value of each of the M categories in the N determining results; and updating the initial image generator based on a difference between the determining result and the real result, an eigenvalue of the feature, and the average probability value.
19. A computer-readable storage medium, wherein the computer-readable storage medium stores program code to be executed by a device, and the program code comprises the image generation method according to claim 1 .
20. A computer-readable storage medium, wherein the computer-readable storage medium stores program code to be executed by a device, and the program code comprises the sample generation method according to claim 7 .