Reverse neural network for object re-identification
The invention relates to a method comprising receiving, by a neural network, a first image comprising at least one target object; receiving, by the neural network, a second image comprising at least one query object; and determining, by the neural network, whether the query object corresponds to the target object, wherein the neural network comprises a discriminator neural network of a generative adversarial network (GAN). The invention further relates to an apparatus and a computer program product that perform the method.
1. A method comprising:
receiving, by a neural network, a first image comprising at least one target object, wherein the neural network comprises a discriminator neural network of a generative adversarial network (GAN);
extracting a characterizing feature from the at least one target object by removing visual content of the feature;
replacing the visual content with visual noise or a visual pattern to generate an estimate of the first image having at least one generated object;
receiving, by the neural network, a second image comprising at least one query object, the at least one query object comprising the at least one target object captured in a manner different from the at least one target object of the first image; and
training the discriminator neural network to discriminate between the second image comprising the at least one query object and the estimate of the first image having the at least one generated object based upon a determined correspondence to the first image.
2. The method according to claim 1 , wherein the second image comprises at least a portion of a current video frame and the first image comprises at least a portion of a previous video frame.
3. The method according to claim 2 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the previous video frame and to provide an estimate of the current video frame to the discriminator neural network.
4. The method according to claim 3 , wherein the modification of the previous video frame comprises extracting at least one characterizing feature from the previous video frame.
5. The method according to claim 4 , wherein the characterizing feature is at least one of a nose, eyes or lips of a face.
6. The method according to claim 3 , wherein the generator neural network is trained by a second discriminator neural network configured to determine whether the estimate of the current video frame generated by the generator neural network is realistic or fake.
7. The method according to claim 1 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the first image.
8. The method according to claim 1 , wherein the at least one query object comprises the at least one target object captured from a location different from a location at which the at least one target object of the first image was captured, by a capture device different from a capture device with which the at least one target object of the first image was captured, at a time different from a time the at least one target object of the first image was captured, or a combination thereof.
9. An apparatus comprising at least one processor and memory including computer program code, the memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform:
receive, by a neural network, a first image comprising at least one target object, wherein the neural network comprises a discriminator neural network of a generative adversarial network (GAN);
extract a characterizing feature from the at least one target object by removing visual content of the feature;
replace the visual content with visual noise or a visual pattern to generate an estimate of the first image having at least one generated object;
receive, by the neural network, a second image comprising at least one query object, the at least one query object comprising the at least one target object captured in a manner different from the at least one target object of the first image; and
train the discriminator neural network to discriminate between the second image comprising the at least one query object and the estimate of the first image having the at least one generated object based upon a determined correspondence to the first image.
10. The apparatus according to claim 9 , wherein the second image comprises at least a portion of a current video frame and the first image comprises at least a portion of a previous video frame.
11. The apparatus according to claim 10 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the previous video frame and to provide an estimate of the current video frame to the discriminator neural network.
12. The apparatus according to claim 11 , wherein the modification of the previous video frame comprises extracting at least one characterizing feature from the previous video frame.
13. The apparatus according to claim 12 , wherein the characterizing feature is at least one of a nose, eyes or lips of a face.
14. The apparatus according to claim 11 , wherein the generator neural network is trained by a second discriminator neural network configured to determine whether the estimate of the current video frame generated by the generator neural network is realistic or fake.
15. The apparatus according to claim 9 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the first image.
16. A computer program product embodied on a non-transitory computer readable medium, comprising computer program code configured to, when executed on at least one processor, cause an apparatus or a system to perform:
receive, by a neural network, a first image comprising at least one target object, wherein the neural network comprises a discriminator neural network of a generative adversarial network (GAN);
extract a characterizing feature from the at least one target object by removing visual content of the feature;
replace the visual content with visual noise or a visual pattern to generate an estimate of the first image having at least one generated object;
receive, by the neural network, a second image comprising at least one query object, the at least one query object comprising the at least one target object captured in a manner different from the at least one target object of the first image; and
train the discriminator neural network to discriminate between the second image comprising the at least one query object and the estimate of the first image having the at least one generated object based upon a determined correspondence to the first image.
17. The computer program product according to claim 16 , wherein the second image comprises at least a portion of a current video frame and the first image comprises at least a portion of a previous video frame.
18. The computer program product according to claim 17 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the previous video frame and to provide an estimate of the current video frame to the discriminator neural network.
19. The computer program product according to claim 18 , wherein the modification of the previous video frame comprises extracting at least one characterizing feature from the previous video frame.
20. The computer program product according to claim 19 , wherein the characterizing feature is at least one of a nose, eyes or lips of a face.
21. The computer program product according to claim 16 , wherein the discriminator neural network is trained by using a generator neural network configured to receive a modified version of the first image.