文章预览
本期带来上周在arXiv公开的论文9篇,涉及3D生成。 一、3D生成 1.[2024-arXiv] MeshAnything V2: Artist-Created Mesh Generation With Adjacent Mesh Tokenization 作者机构:Yiwen Chen, et al. NTU, etc. 论文地址: https://arxiv.org/pdf/2408.02555 项目地址: https://buaacyw.github.io/meshanything-v2/ 贡献: We introduce a novel mesh tokenization method, named Adjacent Mesh Tokenization (AMT). Compared to previous tokenization methods, AMT requires approximately half the token sequence length to represent the same mesh, thereby fundamentally reducing the computational burden of Artist-Created Mesh generation. We equip MeshAnything with AMT and introduce MeshAnything V2. V2 doubles the maximum number of faces that can be generated while achieving significantly better accuracy and efficiency. Extensive experiments show that AMT significantly improves the efficiency and performance of mesh generation. 2.[2024-MM] A General Framework to Boost 3D GS Initializatio
………………………………