Closed duhau closed 3 years ago
Thanks for your interest on our work and sorry for the late reply.
We think it is possible to adapt MeshNet for segmentation task as classification on each face, since we extract semantic features of faces in the model classification task. Roughly speaking, you should prepare segmentation labels for mesh faces, and conduct the similar preprocessing that similifies mesh models and reorganizes the data format. Then modify the network architecture, remove the global pooling and apply an classification module to face features.
I want to know whether this method can be used for model segmentation, if so, how should I prepare for input dates.