Meta-learning for facial recognition
Methods and systems are provided for generating a facial recognition system. A facial recognition system can be implemented using a meta-model based on a trained neural network. A neural network can be trained as multiple classifiers that identify individuals using a small number of images of the individual's face. A meta-model can learn from the neural networks to be capable to identify an individual based on a small number of images. In this way, the facial recognition system uses the meta-model that learns from the neural network trained to identify an individual based on a small number of images. Such a facial recognition system is tested to determine any misidentification for fine-tuning the system. A facial recognition system implemented using such a meta-model is capable of adapting the model to learn identities entered into the system using only a small number of images to enroll an identity into the system.
1. One or more non-transitory computer-readable media having a plurality of executable instructions embodied thereon, which, when executed by one or more processors, cause the one or more processors to perform a method, the method comprising:
generating a meta-model based on a neural network trained using multiple binary classifiers for identifying individuals by classifying overall facial images of the individuals, wherein each of the binary classifiers is trained using a first facial image of one of the individuals, wherein a first binary classifier is trained using one facial image for a first individual to learn an identity for the first individual;
updating the meta-model by validating the meta-model to update for error in the meta-model, the error based on an output identity by the meta-model using a second facial image for the first individual input into the meta-model; and
utilizing the meta-model to identify a second individual based on a new facial image.
2. The media of claim 1 , the method further comprising:
enrolling the second individual in a facial recognition system by inputting one image of the second individual into the facial recognition system such that the meta-model learns to classify the second individual.
3. The media of claim 1 , the method further comprising:
testing the meta-model utilizing a third individual, the third individual different from the classifiers for identifying individuals for which the neural network is trained.
4. The media of claim 3 , wherein testing the facial recognition system utilizes an unknown facial image of the third individual to fine-tune the meta-model based on further error.
5. The media of claim 1 , wherein training the neural network further comprises:
during a first iteration, using the one facial image for the first individual, determining a first loss using a comparison between a first output and a first ground-truth output; and
adjusting the neural network based on the first loss.
6. The media of claim 5 , wherein training the neural network further comprises:
during a second iteration, using a third facial image for the first individual, determining a second loss using a comparison between a second output and a second ground-truth output; and
adjusting the neural network based on the second loss.
7. The media of claim 1 , wherein the error in the meta-model is determined using a binary entropy loss function.
8. The media of claim 1 , further comprising:
further updating the meta-model, wherein the meta-model learns from additional trained iterations of the neural network;
testing the updated meta-model; and
repeating the further updating and the testing until output of the meta-model reaches a desired threshold minimum of loss.
9. A computer-implemented method for training a facial recognition system, the method comprising:
selecting an image with an individual to identify;
inputting the selected image into a trained facial recognition system implemented using a meta-model based on a neural network trained using multiple classifiers for identifying individuals by classifying overall facial images of the individuals, wherein each of the classifiers is trained using a single facial image of one of the individuals, wherein training the facial recognition system includes:
generating the meta-model based on the neural network, wherein the neural network is trained using a first classifier trained using one facial image for a first individual to learn an identity for the first individual,
updating the meta-model by validating the meta-model to update for error in the meta-model, the error based on an output identity by the meta-model using a second facial image for the first individual input into the meta-model, and
testing the meta-model to fine-tune the meta-model by for updating further error; and
outputting an identity for the individual in the image using the trained facial recognition system.
10. The computer-implemented method of claim 9 , wherein the identity of the individual in the image is displayed via a user device.
11. The computer-implemented method of claim 9 , further comprising:
enrolling an individual in the facial recognition system by inputting one image of the individual into the facial recognition system such that the meta-model learns an identity of the individual.
12. The computer-implemented method of claim 9 , wherein the testing of the facial recognition system further comprises utilizing an unknown facial image as input into the meta-model to determine the further error.
13. The computer-implemented method of claim 9 , wherein training the neural network further comprises:
during a first iteration, using the single facial image for the first individual, determining a first loss using a comparison between a first output and a first ground-truth output; and
adjusting the neural network based on the first loss.
14. The computer-implemented method of claim 13 , the training further comprising:
during a second iteration, using a third facial image for the first individual, determining a second loss using a comparison between a second output and a second ground-truth output; and
adjusting the neural network based on the second loss.
15. The computer-implemented method of claim 9 , wherein training the facial recognition system further comprises:
further updating the meta-model, wherein the meta-model learns from additional training iterations of the neural network;
further testing the facial recognition system implemented using the updated meta-model;
repeating the further updating and further testing until output of the facial recognition system reaches a desired threshold minimum of loss.
16. The computer-implemented method of claim 9 , wherein the error in the meta-model is determined using a binary entropy loss function.
17. A computing system comprising:
means for training a facial recognition system, wherein the facial recognition system is implemented using a meta-model, the meta-model generated based on a neural network trained using multiple binary classifiers for identifying individuals by classifying overall facial images of the individuals, wherein each of the multiple binary classifiers is trained using a single facial image of one of the individuals; and
means for identifying an individual in an image using the trained facial recognition system, the trained neural network system implemented using the meta-model.
18. The system of claim 17 , wherein the instructions further cause the computing system to provide:
means for updating the meta-model, wherein the meta-model learns from additional training iterations of the neural network.
19. The system of claim 17 , wherein the instructions further cause the computing system to provide:
means for testing the meta-model to fine-tune the meta-model.
20. The system of claim 17 , wherein the instructions further cause the computing system to provide:
means for outputting the identity of the individual in the image.