Hello. Im new at Motion Capture field. Currently, I have a dance motion data. For now, Im quite having a trouble to decide which method is the best for retarget a dance motion data. Since Im quite not familiar with "Deep Learning" method, I want to ask the researchers here, is it the best method to implement? Or is it another way around? Thank you
Yes it is: see & watch for example:https://openreview.net/pdf?id=HknqPIJDM
https://www.youtube.com/watch?v=fR4h4OjZSdU
https://www.youtube.com/watch?v=Kv2ayFELxHg
Muhammad Ali My friend suggests me to use the tools provided by the Autodesk Maya, just add some tweaked code. He also told me that deep learning needs a lot of machine learning knowledge. Is is true? What is your opinion on my friend's statement?
Deep learning is at the moment one the best methods for analyzing structured data. In your case that will be motion capture data which means analyzing 4D tensors (3D image in time). The challenge will not be the coding part for you since there are multiple softwares in which deep learning based models are implented (look for object detection plugings in ImageJ, Matlab and ...). The difficulty will actually come from providing the sufficient annotated dataset that deep learning models need to achieve high accuracy. So, I sugggest, if you have access to enough annotated data, as a CS student, don't be overwhelmed by the coding complexity of DL and try using it. Otherwise, you could also try applying of the shelf trained DL models which depending on your data (and similarity to the data the models are trained on) might achieve high accuracy in prediction.
- Badges
- Science topic