Universal feature representation learning for face recognition
A computer-implemented method for implementing face recognition includes receiving training data including a plurality of augmented images each corresponding to a respective one of a plurality of input images augmented by one of a plurality of variations, splitting a feature embedding generated from the training data into a plurality of sub-embeddings each associated with one of the plurality of variations, associating each of the plurality of sub-embeddings with respective ones of a plurality of confidence values, and applying a plurality of losses including a confidence-aware identification loss and a variation-decorrelation loss to the plurality of sub-embeddings and the plurality of confidence values to improve face recognition performance by learning the plurality of sub-embeddings.
1. A computer-implemented method for implementing face recognition, comprising:
receiving training data including a plurality of augmented images each corresponding to a respective one of a plurality of input images augmented by one of a plurality of variations;
splitting a feature embedding generated from the training data into a plurality of sub-embeddings each associated with one of the plurality of variations;
associating each of the plurality of sub-embeddings with respective ones of a plurality of confidence values; and
applying a plurality of losses including a confidence-aware identification loss and a variation-decorrelation loss to the plurality of sub-embeddings and the plurality of confidence values to improve face recognition performance by learning the plurality of sub-embeddings.
2. The method as recited in claim 1 , further comprising receiving the plurality of input images, and generating the plurality of augmented images by applying online variation data augmentation to introduce the plurality of variations into the plurality of input images.
3. The method as recited in claim 1 , wherein the plurality of variations include blurriness, occlusion and pose.
4. The method as recited in claim 1 , where the confidence-aware identification loss incorporates a confidence score associated with a loss margin to narrow a within-identity distribution and widen an inter-identity distribution.
5. The method as recited in claim 1 , wherein the variation-decorrelation loss applies a classification loss on the plurality of variations and an adversarial loss on different partitions of the plurality of sub-embeddings.
6. The method as recited in claim 1 , wherein the plurality of losses further includes an additional regularization to constrain confidence from growing arbitrary large.
7. The method as recited in claim 1 , further comprising obtaining pairwise similarity scores according to respective image pairs, each of the pairwise similarity scores corresponding to learned sub-embedding importance.
8. A computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method for implementing face recognition, the method performed by the computer comprising:
receiving training data including a plurality of augmented images each corresponding to a respective one of a plurality of input images augmented by one of a plurality of variations;
splitting a feature embedding generated from the training data into a plurality of sub-embeddings each associated with one of the plurality of variations;
associating each of the plurality of sub-embeddings with respective ones of a plurality of confidence values; and
applying a plurality of losses including a confidence-aware identification loss and a variation-decorrelation loss to the plurality of sub-embeddings and the plurality of confidence values to improve face recognition performance by learning the plurality of sub-embeddings.
9. The computer program product as recited in claim 8 , wherein the method further includes receiving the plurality of input images, and generating the plurality of augmented images by applying online variation data augmentation to introduce the plurality of variations into the plurality of input images.
10. The computer program product as recited in claim 8 , wherein the plurality variations of include blurriness, occlusion and pose.
11. The computer program product as recited in claim 8 , where the confidence-aware identification loss incorporates a confidence score associated with a loss margin to narrow a within-identity distribution and widen an inter-identity distribution.
12. The computer program product as recited in claim 8 , wherein the variation-decorrelation loss applies a classification loss on the plurality of variations and an adversarial loss on different partitions of the plurality of sub-embeddings.
13. The computer program product as recited in claim 8 , wherein the plurality of losses further includes an additional regularization to constrain confidence from growing arbitrary large.
14. The computer program product as recited in claim 8 , wherein the method further includes obtaining pairwise similarity scores according to respective image pairs, each of the pairwise similarity scores corresponding to learned sub-embedding importance.
15. A system for implementing face recognition, comprising:
a memory device storing program code; and
at least one processor device operatively coupled to the memory device and configured to execute program code stored on the memory device to:
receive training data including a plurality of augmented images each corresponding to a respective one of a plurality of input images augmented by one of a plurality of variations;
split a feature embedding generated from the training data into a plurality of sub-embeddings each associated with one of the plurality of variations;
associate each of the plurality of sub-embeddings with respective ones of a plurality of confidence values; and
apply a plurality of losses including a confidence-aware identification loss and a variation-decorrelation loss to the plurality of sub-embeddings and the plurality of confidence values to improve face recognition performance by learning the plurality of sub-embeddings.
16. The system as recited in claim 15 , wherein the system further includes receiving the plurality of input images, and generating the plurality of augmented images by applying online variation data augmentation to introduce the plurality of variations into the plurality of input images, and wherein the plurality variations of include blurriness, occlusion and pose.
17. The system as recited in claim 15 , where the confidence-aware identification loss incorporates a confidence score associated with a loss margin to narrow a within-identity distribution and widen an inter-identity distribution.
18. The system as recited in claim 15 , wherein the variation-decorrelation loss applies a classification loss on the plurality of variations and an adversarial loss on different partitions of the plurality of sub-embeddings.
19. The system as recited in claim 15 , wherein the plurality of losses further includes an additional regularization to constrain confidence from growing arbitrary large.
20. The system as recited in claim 15 , wherein the system further includes obtaining pairwise similarity scores according to respective image pairs, each of the pairwise similarity scores corresponding to learned sub-embedding importance.