@techreport{oai:ipsj.ixsq.nii.ac.jp:00233208,
 author = {Yuhao, Dou and Tomohiko, Mukai and Yuhao, Dou and Tomohiko, Mukai},
 issue = {7},
 month = {Mar},
 note = {This research proposes an unsupervised facial retargeting method based on graph convolution networks and a variant of generative adversarial networks. Our insight is that animations of different face models can be represented by a common set of landmark points. To enable the down-sampling and up-sampling on the graph structure, we introduce graph pooling and graph unpooling using the landmark set. In the encoding stage, the graph pooling operator merges the several adjacent node features into a landmark node to compose the shared latent space. The latent code is decoded to reconstruct the face model by the graph unpooling in the decoding stage. Our approach enables retargeting animation between face models of different mesh topologies with unpaired data in the common latent space., This research proposes an unsupervised facial retargeting method based on graph convolution networks and a variant of generative adversarial networks. Our insight is that animations of different face models can be represented by a common set of landmark points. To enable the down-sampling and up-sampling on the graph structure, we introduce graph pooling and graph unpooling using the landmark set. In the encoding stage, the graph pooling operator merges the several adjacent node features into a landmark node to compose the shared latent space. The latent code is decoded to reconstruct the face model by the graph unpooling in the decoding stage. Our approach enables retargeting animation between face models of different mesh topologies with unpaired data in the common latent space.},
 title = {Human Facial Animation Retargeting by Unsupervised Neural Network},
 year = {2024}
}