Methods and systems for interpolation of disparate inputs
Systems and methods are provided for interpolation of disparate inputs. A radial basis function neural network (RBFNN) may be used to interpolate the pose of a digital character. Input parameters to the RBFNN may be separated by data type (e.g. angular vs. linear) and manipulated within the RBFNN by distance functions specific to the data type (e.g. use an angular distance function for the angular input data). A weight may be applied to each distance to compensate for input data representing different variables (e.g. clavicle vs. shoulder). The output parameters of the RBFNN may be a set of independent values, which may be combined into combination values (e.g. representing x, y, z, w angular value in SO(3) space).
1. A method comprising:
receiving input data comprising at least one angular component and at least one linear component;
providing the input data as input to at least one neural network (NN) that has been trained to evaluate the at least one angular component differently than the at least one linear component; and
receiving output data that is generated by the at least one NN based on the different evaluation of the at least one angular component and the at least one linear component,
wherein the input data represents a low-order skeleton of a digital character, and the output data represents a high-order skeleton of a digital character.
2. The method of claim 1 , wherein at least one of the neural networks is a feed-forward neural network (FFNN).
3. The method of claim 2 , wherein the at least one FFNN is a fully connected network.
4. The method of claim 3 , wherein the FFNN comprises a single hidden layer.
5. The method of claim 4 , wherein the FFNN comprises rectified linear unit activation functions.
6. The method of claim 5 , wherein the hidden layer is a residual NN block.
7. The method of claim 1 , wherein at least one of the NNs is a radial basis function neural network (RBFNN).
8. The method of claim 1 , wherein the input data describes a pose of a digital character.
9. The method of claim 1 , wherein the output data describes a pose of a digital character.
10. The method of claim 1 , wherein one or more of the input data and the output data further comprises a third component.
11. The method of claim 10 , wherein the angular, linear, and third components are each different components of motion.
12. The method of claim 11 , wherein:
the at least one angular component describes a rotational motion;
the at least one linear component describes a translational motion; and
the third component describes scale.
13. The method of claim 10 , wherein the at least one NN comprises a first FFNN that evaluates the at least one angular component and a second FFNN that evaluates the at least one linear component.
14. The method of claim 13 , wherein the at least one NN comprises a third FFNN that evaluates the third component.
15. The method of claim 1 , wherein the at least one NN comprises a plurality of sample nodes, wherein each sample node corresponds to a training pose, and wherein at least one of the training poses comprises at least one angular and linear component.
16. The method of claim 1 , wherein the at least one NN evaluates the at least one angular component differently than the at least one linear component by evaluating the at least one angular component in special orthogonal group in three dimensions (SO(3)) space and by evaluating the at least one linear component utilizing a Euclidean distance formula.
17. The method of claim 1 , wherein the at least one angular component is in an orthogonal group in three dimensional space, the orthogonal group assigned a weight that represents a relative contribution to an overall movement of a digital character.
18. A method comprising:
receiving input data comprising at least one angular component and at least one linear component;
providing the input data as input to at least one neural network (NN) that has been trained to evaluate the at least one angular component differently than the at least one linear component;
receiving output data that is generated by the at least one NN based on the different evaluation of the at least one angular component and the at least one linear component, and
training the at least one NN, wherein the training comprises:
receiving training data comprising training input data and training output data, wherein the training input data and the training output data represent one or more training poses, wherein at least one of the one or more training poses comprise an input angular component, an input linear component, an output angular component, and an output linear component;
grouping the input angular components from each of the one or more poses into an input angular component group;
grouping the input linear components from each of the one or more poses into an input linear component group; and
providing the training input data as input to train the at least one NN, wherein the input angular component group is evaluated differently than the input linear component group, wherein the evaluation results in the output angular component and the output linear component.
19. A system comprising:
at least one processor that executes at least one neural network (NN); and
memory communicatively coupled to the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving input data comprising at least one angular component and at least one linear component;
providing the input data as input to at least one neural network (NN) that has been trained to evaluate the at least one angular component differently than the at least one linear component; and
receiving output data that is generated by the at least one NN based on the different evaluation of the at least one angular component and the at least one linear component,
wherein the input data represents a low-order skeleton of a digital character, and the output data represents a high-order skeleton of a digital character.