The next-coordinate prediction paradigm has emerged as the de facto standard in current auto-regressive mesh generation methods.Despite their effectiveness, there is no efficient measurement for the various tokenizers that serialize meshes into sequences. In this paper, we introduce a new metric Per-Token-Mesh-Entropy (PTME) to evaluate the existing mesh tokenizers theoretically without any training.Building upon PTME, we propose a plug-and-play tokenization technique called coordinate merging.It further improves the compression ratios of existing tokenizers by rearranging and merging the most frequent patterns of coordinates.Through experiments on various tokenization methods like MeshXL, MeshAnything V2, and Edgerunner, we further validate the performance of our method.We hope that the proposed PTME and coordinate merging can enhance the existing mesh tokenizers and guide the further development of native mesh generation.

AI models generate 3D meshes by first converting them into sequences of tokens. A major challenge is efficiently evaluating how well these mesh tokenizers work, as it usually requires training the full AI model. We introduce a new, training-free metric called Per-Token-Mesh-Entropy (PTME) that quickly measures how effectively geometric information is packed into the mesh sequence tokens. Using PTME, we also propose a method to improve existing tokenizers by rearranging and merging frequent patterns within the mesh sequence, creating a more compressed representation. Our PTME metric and the mesh sequence merging technique enable researchers to rapidly evaluate and enhance mesh tokenization methods, helping AI models generate more detailed and higher-quality meshes for applications in computer graphics and design.