Biometric task network
Output can be provided from a selected biometric analysis task that is one of a plurality of biometric analysis tasks based on an image provided from an image sensor. The selected biometric analysis task can be performed in a deep neural network that includes a common feature extraction neural network, a plurality of biometric task-specific neural networks and a plurality of expert pooling neural networks that perform the plurality of biometric analysis tasks by inputting the image to the common feature extraction network to determine latent variables. The latent variables can be input to the plurality of biometric task-specific neural networks to determine a plurality of first outputs. Concatenated first output results can be formed and the concatenated plurality of first result outputs and the latent variables can be input to the plurality of expert pooling neural networks to determine one or more biometric analysis task outputs.
1. A system, comprising:
a computer that includes a processor and a memory, the memory including instructions executable by the processor to:
receive an image at a common feature extraction network to determine skip connected latent variables;
receive the skip connected latent variables at a plurality of biometric task-specific neural networks to determine a plurality of first outputs;
receive the first outputs and the skip connected latent variables at a plurality of expert pooling neural networks to determine one or more biometric analysis task outputs; and
wherein the selected one or more biometric analysis task outputs are determined in a deep neural network that includes the common feature extraction neural network, the plurality of biometric task-specific neural networks and the plurality of expert pooling networks that are trained based on a joint loss function which compensates for differences in sizes of training datasets for the biometric task-specific neural networks.
2. The system of claim 1 , the instructions including further instructions to operate a device based on the selected biometric analysis task output.
3. The system of claim 1 , wherein the plurality of biometric analysis tasks includes biometric identification, liveliness determination, drowsiness determination, gaze determination, pose determination, and facial feature segmentation.
4. The system of claim 1 , wherein the common feature extraction neural network includes a plurality of convolutional layers.
5. The system of claim 1 , wherein the plurality of biometric task-specific neural networks includes a plurality of fully connected layers.
6. The system of claim 1 , wherein the plurality of expert pooling neural networks includes a plurality of fully connected layers.
7. The system of claim 1 , wherein the one or more of first outputs from the plurality of biometric task-specific neural networks are input to a SoftMax function before being concatenated.
8. The system of claim 1 , wherein one or more of the outputs from the plurality of expert pooling neural networks are input to a SoftMax function before being output as biometric analysis task outputs.
9. The system of claim 1 , wherein one or more of the biometric analysis task outputs are set to zero during training.
10. The system of claim 1 , the instructions including further instructions to train the deep neural network by combining the biometric analysis task outputs with ground truth to determine a joint loss function.
11. The system of claim 10 , wherein the joint loss function is backpropagated through the deep neural network to determine weights.
12. The system of claim 1 , wherein the deep neural network is configured to include the common feature extraction network and a subset of the biometric task-specific neural networks during inference based on a selected biometric analysis task.
13. A method, comprising:
receiving an image at a common feature extraction network to determine skip connected latent variables;
receiving the skip connected latent variables at a plurality of biometric task-specific neural networks to determine a plurality of first outputs;
receiving the first outputs and the skip connected latent variables at a plurality of expert pooling neural networks to determine one or more biometric analysis task outputs; and
wherein the selected one or more biometric analysis task outputs are determined in a deep neural network that includes the common feature extraction neural network, the plurality of biometric task-specific neural networks and the plurality of expert pooling networks that are trained based on a joint loss function which compensates for differences in sizes of training datasets for the biometric task-specific neural networks.
14. The method of claim 13 , further comprising operating a device based on the selected biometric analysis task output.
15. The method of claim 13 , wherein the plurality of biometric analysis tasks includes biometric identification, liveliness determination, drowsiness determination, gaze determination, pose determination, and facial feature segmentation.
16. The method of claim 13 , wherein the common feature extraction neural network includes a plurality of convolutional layers.
17. The method of claim 13 , wherein the plurality of biometric task-specific neural networks includes a plurality of fully connected layers.
18. The method of claim 13 , wherein the plurality of expert pooling neural networks includes a plurality of fully connected layers.
19. The method of claim 13 , wherein the one or more of first outputs from the plurality of biometric task-specific neural networks are input to a SoftMax function before being concatenated.
20. The method of claim 13 , wherein one or more of the outputs from the plurality of expert pooling neural networks are input to a SoftMax function before being output as biometric analysis task outputs.