Motion synthesis apparatus and motion synthesis method
A motion synthesis motion synthesis method including: obtaining, by a motion synthesis apparatus, content feature values and style feature values according to content motion data and style motion data; generating, by the motion synthesis apparatus, target feature values using the obtained content feature values and style feature values; recognizing, by the motion synthesis apparatus, synthesized motion data and obtaining synthesized motion feature values from the recognized synthesized motion data; and obtaining, by the motion synthesis apparatus, loss by using the synthesized motion feature values and the target feature values and updating the synthesized motion data according to the obtained loss.
1. A motion synthesis method comprising:
obtaining, by a motion synthesis apparatus, content feature values and style feature values according to content motion data and style motion data of an object;
generating, by the motion synthesis apparatus, target feature values using the obtained content feature values and style feature values;
recognizing, by the motion synthesis apparatus, synthesized motion data and obtaining synthesized motion feature values from the recognized synthesized motion data; and
obtaining, by the motion synthesis apparatus, loss by using the synthesized motion feature values and the target feature values and updating the synthesized motion data according to the obtained loss,
wherein the generating the target feature values comprising generating a vector for minimizing an error in a value obtained by adding a content loss and a weighted style loss, the content loss indicating a difference between the target feature values and the content feature values and the weighted style loss indicating a difference between the target feature values and the style feature values, to which a weight is applied.
2. The motion synthesis method of claim 1 , wherein the updating of the synthesized motion data comprises:
using a back-propagation algorithm until the synthesized motion feature values and the target feature values are matched.
3. The motion synthesis method of claim 1 , wherein the motion synthesis apparatus obtains feature values using an untrained convolutional neural network.
4. The motion synthesis method of claim 1 , wherein the content motion data and the style motion data are animation data.
5. The motion synthesis method of claim 1 , wherein the content motion data and the style motion data comprise information about a bone.
6. The motion synthesis method of claim 1 , wherein the content loss and the weighted style loss are respectively indexed according to each part of the object.
7. The motion synthesis method of claim 6 , wherein a different value of the weight is assigned according to each part of the object in which each of the weighted style loss is indexed.
8. The motion synthesis method of claim 1 , wherein the motion synthesis apparatus obtains the content feature values and the style feature values by providing the content motion data and the style motion data as an input to one convolutional neural network.
9. A motion synthesis apparatus, the motion synthesis apparatus comprising a processor configured to:
obtain content feature values and style feature values according to content motion data and style motion data of an object;
generate target feature values using the obtained content feature values and style feature values;
recognize synthesized motion data and obtains synthesized motion feature values from the recognized synthesized motion data; and
obtain loss by using the synthesized motion feature values and the target feature values and updates the synthesized motion data according to the obtained loss,
wherein the processor is further configured to generate the target feature values by generating a vector for minimizing an error in a value obtained by adding a content loss and a weighted style loss, the content loss indicating a difference between the target feature values and the content feature values and the weighted style loss indicating a difference between the target feature values and the style feature values, to which a weight is applied.
10. The motion synthesis apparatus of claim 9 , wherein the updating of the synthesized motion data comprises:
using a back-propagation algorithm until the synthesized motion feature values and the target feature values are matched.
11. The motion synthesis apparatus of claim 9 , wherein the processor is configured to obtain feature values using an untrained convolutional neural network.
12. The motion synthesis apparatus of claim 9 , wherein the content motion data and the style motion data are animation data.
13. The motion synthesis apparatus of claim 9 , wherein the content motion data and the style motion data comprise information about a bone.