Face reconstruction from a learned embedding
The present disclosure provides systems and methods that perform face reconstruction based on an image of a face. In particular, one example system of the present disclosure combines a machine-learned image recognition model with a face modeler that uses a morphable model of a human's facial appearance. The image recognition model can be a deep learning model that generates an embedding in response to receipt of an image (e.g., an uncontrolled image of a face). The example system can further include a small, lightweight, translation model structurally positioned between the image recognition model and the face modeler. The translation model can be a machine-learned model that is trained to receive the embedding generated by the image recognition model and, in response, output a plurality of facial modeling parameter values usable by the face modeler to generate a model of the face.
1. A computer-implemented method to perform machine learning on images of a face, the method comprising:
obtaining, by one or more computing devices, a set of training data that comprises an image of a face and a first set of facial modeling parameter values associated with a three-dimensional model of the face;
inputting, by the one or more computing devices, the image of the face into a face reconstruction system comprising one or more machine-learned neural networks;
receiving, by the one or more computing devices, a second set of facial modeling parameter values generated by the face reconstruction system based at least in part on the image of the face;
evaluating, by the one or more computing devices, a loss function that compares the first set of facial modeling parameter values and the second set of facial modeling parameter values; and
training, by the one or more computing devices, the one or more machine-learned convolutional neural networks of the face reconstruction system based at least in part on the loss function.
2. The computer-implemented method of claim 1 , wherein the second set of facial modeling parameter values describe facial attributes of the face.
3. The computer-implemented method of claim 1 , wherein the face reconstruction system comprises:
a machine-learned convolutional neural network configured to receive and process the image of the face to generate an embedding for the face; and
a second neural network configured to receive and process the embedding for the face to generate the second set of facial modeling parameter values for the face;
wherein training, by the one or more computing devices, the one or more machine-learned neural networks of the face reconstruction system based at least in part on the loss function comprises training at least the second neural network based at least in part on the loss function.
4. The computer-implemented method of claim 3 , wherein the second neural network comprises a regression neural network.
5. The computer-implemented method of claim 3 , wherein the second neural network comprises one or more fully connected layers.
6. The computer-implemented method of claim 1 , wherein the first set of facial modeling parameters are generated from the image of the face by a computer-implemented facial modeling system.
7. The computer-implemented method of claim 1 , wherein the image of the face comprises a controlled image of the face with front-facing pose.
8. The computer-implemented method of claim 1 , wherein the second set of facial modeling parameter values are usable by a face modeler to generate a three-dimensional mesh model of the face.
9. The computer-implemented method of claim 1 , further comprising:
generating, by the one or more computing devices, a three-dimensional mesh model of the face based on the second set of facial modeling parameter values.
10. The computer-implemented method of claim 1 , wherein the second set of facial modeling parameter values describe landmark positions for the face.