LoST introduces a semantic-first 3D tokenizer that reduces the token count for 3D shape generation by up to 99.9%.
March 19, 2026
Original Paper
LoST: Level of Semantics Tokenization for 3D Shapes
arXiv · 2603.17995
The Takeaway
Autoregressive 3D generation is typically bottlenecked by the massive number of tokens required for geometric detail. By prioritizing semantic salience over raw geometry, LoST enables high-quality 3D shape synthesis and reconstruction with a fraction of the compute and memory overhead of previous SOTA models.
From the abstract
Tokenization is a fundamental technique in the generative modeling of various modalities. In particular, it plays a critical role in autoregressive (AR) models, which have recently emerged as a compelling option for 3D generation. However, optimal tokenization of 3D shapes remains an open question. State-of-the-art (SOTA) methods primarily rely on geometric level-of-detail (LoD) hierarchies, originally designed for rendering and compression. These spatial hierarchies are often token-inefficient