QuadLink: Autoregressive Quad-Dominant Mesh Generation via Point-Relation Learning

Researchers have introduced three new frameworks for generating 3D meshes: QuadLink, MeshFlow, and MeshWeaver. These frameworks aim to improve the generation of quad-dominant meshes from point clouds.

QuadLink, introduced by researchers on May 16, 2026^[1], is a unified framework for quad-dominant mesh generation. It formulates polygonal mesh generation as a hybrid centroid-conditioned vertex linking model, enabling the generation of sparse and anisotropic quad-dominant meshes. QuadLink achieves improved geometric fidelity and topological quality compared to prior baselines. Meanwhile, MeshFlow, introduced on arXiv on June 2, 2026^[2], uses a Variational Autoencoder (VAE) to represent both continuous vertex positions and discrete connectivity in a continuous latent space. MeshFlow generates meshes 18x faster than the fastest AR generator while achieving excellent accuracy across standard mesh-generation metrics. Additionally, researchers have introduced MeshWeaver^[3], an autoregressive framework for mesh generation that treats mesh generation as a surface weaving process. MeshWeaver achieves a state-of-the-art compression ratio of 18%^[3] and can generate meshes with up to 16K faces. MeshWeaver also significantly improves geometric fidelity over prior approaches.