Implemented human to game character transfer based on the UCB paper with the same name. Used pose estimation api by CMU and trained a Conditional GAN model on the pose estimated stick pose stick figure to learn a video game character. Further refinement in the project can be of great use to get animated version of real-life movies.
HUMAN TO HUMAN TRANSFER
Training:
Initial Training To see how is GAN learning the images:
Transfer Human to Human: We transfer from a freestyle technique from a freestyle dancer on a Zumba dancer Sample
Link for the video(to show transfer): https://drive.google.com/open?id=1F9ijYSzb_F4c47Et5K2-zy6EXJ6Gyetk
References [1]Everybody dance now: https://arxiv.org/abs/1808.07371 [2]Image-to-Image Translation with Conditional Adversarial Networks: https://arxiv.org/pdf/1611.07004.pdf