[1Win4-87] Learning and Generation of Motions using Transformer
Keywords:Deep Learning, Motion, Generation
In this study, we aimed to generate output motions corresponding to input motions using deep learning. We constructed a motion generation model and evaluated the generated motions.
For motion data acquisition, we used Azure Kinect DK, and for the deep learning model, we adopted a Transformer-based time-series prediction model.
In the evaluation experiment, we generated motions using the trained model and conducted a questionnaire survey in which participants visually assessed the quality of the generated motions.
As a result, the generated motions were generally perceived as natural human movements. However, for some participants, it was found that the generated motions did not accurately correspond to their own movements.
Finally, we discussed the results of this evaluation experiment.
For motion data acquisition, we used Azure Kinect DK, and for the deep learning model, we adopted a Transformer-based time-series prediction model.
In the evaluation experiment, we generated motions using the trained model and conducted a questionnaire survey in which participants visually assessed the quality of the generated motions.
As a result, the generated motions were generally perceived as natural human movements. However, for some participants, it was found that the generated motions did not accurately correspond to their own movements.
Finally, we discussed the results of this evaluation experiment.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.