eduzhai > Applied Sciences > Engineering >

Convolutional Generation of Textured 3D Meshes

  • Save

... pages left unread,continue reading

Document pages: 19 pages

Abstract: While recent generative models for 2D images achieve impressive visualresults, they clearly lack the ability to perform 3D reasoning. This heavilyrestricts the degree of control over generated objects as well as the possibleapplications of such models. In this work, we bridge this gap by leveragingrecent advances in differentiable rendering. We design a framework that cangenerate triangle meshes and associated high-resolution texture maps, usingonly 2D supervision from single-view natural images. A key contribution of ourwork is the encoding of the mesh and texture as 2D representations, which aresemantically aligned and can be easily modeled by a 2D convolutional GAN. Wedemonstrate the efficacy of our method on Pascal3D+ Cars and CUB, both in anunconditional setting and in settings where the model is conditioned on classlabels, attributes, and text. Finally, we propose an evaluation methodologythat assesses the mesh and texture quality separately.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...