NeurIPS 2020

Neural Mesh Flow: 3D Manifold Mesh Generation via Diffeomorphic Flows

Meta Review

3D shape generation with a neural net is studied. After a lengthy discussion all reviewers appreciated the contributions of the paper. This being said, some concerns remained which the reviewers wanted the authors to address (AC concurs): 1. An ablation where the point and shape features are concatenated (instead of dot-multiplied) in the final version. 2. Clarification regarding whether the ODE solver is operating over the linear-layer's 512D response (and not the 3D point directly). If the solver is operating over the 512D response, then the text needs to clarify its claims and relationship to augmented NODE. 3. Inclusion of this reference that appeared at CVPR: Leveraging 2D Data to Learn Textured 3D Mesh Generation. Paul Henderson, Vagia Tsiminaki, Christoph H. Lampert. 4. Merging of the references [8] and [16]. The camera ready should include those points.