Realistic head turns and face animation synthesis on mobile device

1. A method comprising:

receiving, by a computing device, a source frame of a source video, the source frame including a head and a face of a source actor;

generating, by the computing device and based on the source frame, source pose parameters corresponding to a pose of the head of the source actor and a facial expression of the source actor in the source frame of the source video, the source pose parameters including source identity information associated with the face of the source actor;

receiving, by the computing device, a target image including a target head and a target face of a target person;

determining, by the computing device, target identity information associated with the target head and the target face of the target person;

replacing, by the computing device, based on the target image, the source identity information in the source pose parameters with the target identity information to obtain further source pose parameters; and

generating, by the computing device and based on the target identity information and the further source pose parameters, an output frame of an output video, the output frame including a modified image of the target face and the target head, the target face and the target head adopting the pose of the head and the facial expression of the source actor.

2. The method of claim 1 , wherein:

the source identity information includes at least one of the following: a distance between eyes in the face of the source actor and a width of a mouth in the face of the source actor; and

the target identity information includes at least one of the following: a distance between eyes in the target face of the target person and a width of a mouth in the target face of the target person.

3. The method of claim 1 , wherein:

the source pose parameters include two-dimensional facial key points corresponding to the pose of the head of the source actor, the facial expression of the source actor, and the source identity information; and

the further source pose parameters include further two-dimensional facial key points corresponding to the pose of the head of the source actor, the facial expression of the source actor, and the target identity information.

4. The method of claim 3 , wherein the replacing the source identity information in the source pose parameters includes:

fitting a three-dimensional morphable model to the two-dimensional facial key points, the three-dimensional morphable model including source identity parameters associated with the source identity information;

replacing the source identity parameters in the three-dimensional morphable model with target identity parameters associated with the target identity information to obtain a modified three-dimensional morphable model; and

projecting the modified three-dimensional morphable model into a two-dimensional space to obtain the further two-dimensional facial key points.

5. The method of claim 3 , wherein the replacing the source identity information in the source pose parameters is performed by a neural network trained to modify the two-dimensional facial key points to obtain the further two-dimensional facial key points.

6. The method of claim 5 , wherein the neural network is trained based on synthetic data generated with a three-dimensional morphable model.

7. The method of claim 3 , wherein the replacing the source identity information in the source pose parameters includes:

determining, based on the target image and the source frame, a set of affine transformations; and

applying the set of affine transformations to the two-dimensional facial key points to obtain the further two-dimensional facial key points.

8. The method of claim 7 , wherein the set of affine transformations includes an affine transformation for a triangle in a mesh based on coordinates of the two-dimensional facial key points.

9. The method of claim 8 , wherein the set of affine transformations is obtained based on minimization of a difference between the triangle in the mesh and a further tringle in a further mesh, the further mesh being based on coordinates of target facial key points associated with features of the target face and corresponding to the two-dimensional facial key points.

10. The method of claim 1 , wherein the source pose parameters include parameters associated with a parametric facial expression model.

11. A computing device comprising:

a processor; and

a memory storing instructions that, when executed by the processor, configure the computing device to:

receive a source frame of a source video, the source frame including a head and a face of a source actor;

generate, based on the source frame, source pose parameters corresponding to a pose of the head of the source actor and a facial expression of the source actor in the source frame of the source video, the source pose parameters including source identity information associated with the face of the source actor;

receive a target image including a target head and a target face of a target person;

determine target identity information associated with the target head and the target face of the target person;

replace, based on the target image, the source identity information in the source pose parameters with the target identity information to obtain further source pose parameters; and

generate, based on the target identity information and the further source pose parameters, an output frame of an output video, the output frame including a modified image of the target face and the target head, the target face and the target head adopting the pose of the head and the facial expression of the source actor.

12. The computing device of claim 11 , wherein:

the source identity information includes at least one of the following: a distance between eyes in the face of the source actor and a width of a mouth in the face of the source actor; and

the target identity information includes at least one of the following: a distance between eyes in the target face of the target person and a width of a mouth in the target face of the target person.

13. The computing device of claim 11 , wherein:

the source pose parameters include two-dimensional facial key points corresponding to the pose of the head of the source actor, the facial expression of the source actor, and the source identity information; and

the further source pose parameters include further two-dimensional facial key points corresponding to the pose of the head of the source actor, the facial expression of the source actor, and the target identity information.

14. The computing device of claim 13 , wherein the replacing the source identity information in the source pose parameters includes:

fitting a three-dimensional morphable model to the two-dimensional facial key points, the three-dimensional morphable model including source identity parameters associated with the source identity information;

replacing the source identity parameters in the three-dimensional morphable model with target identity parameters associated with the target identity information to obtain a modified three-dimensional morphable model; and

projecting the modified three-dimensional morphable model into a two-dimensional space to obtain the further two-dimensional facial key points.

15. The computing device of claim 13 , wherein the replacing the source identity information in the source pose parameters is performed by a neural network trained to modify the two-dimensional facial key points to obtain the further two-dimensional facial key points.

16. The computing device of claim 15 , wherein the neural network is trained based on synthetic data generated with a three-dimensional morphable model.

17. The computing device of claim 13 , wherein the replacing the source identity information in the source pose parameters includes:

determining, based on the target image and the source frame, a set of affine transformations; and

applying the set of affine transformations to the two-dimensional facial key points to obtain the further two-dimensional facial key points.

18. The computing device of claim 17 , wherein the set of affine transformations includes an affine transformation for a triangle in a mesh based on coordinates of the two-dimensional facial key points.

19. The computing device of claim 18 , wherein the set of affine transformations is obtained based on minimization of a difference between the triangle in the mesh and a further tringle in a further mesh, the further mesh being based on coordinates of target facial key points associated with features of the target face and corresponding to the two-dimensional facial key points.

20. A non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium including instructions that, when executed by a computing device, cause the computing device to:

receive a source frame of a source video, the source frame including a head and a face of a source actor;

generate, based on the source frame, source pose parameters corresponding to a pose of the head of the source actor and a facial expression of the source actor in the source frame of the source video, the source pose parameters including source identity information associated with the face of the source actor;

receive a target image including a target head and a target face of a target person;

determine target identity information associated with the target head and the target face of the target person;

replace, based on the target image, the source identity information in the source pose parameters with the target identity information to obtain further source pose parameters; and

generate, based on the target identity information and the further source pose parameters, an output frame of an output video, the output frame including a modified image of the target face and the target head, the target face and the target head adopting the pose of the head and the facial expression of the source actor.