Newest papers

MixPT: Point Transformer with Interleaved Window and Global Attention

Minkyun Seo, Hyungtae Lim, Jaesik Park · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31455 · cited by 0 works (OpenAlex) · open-access PDF

French

Deux approches communes sont l'attention de fenêtre et l'attention globale. Ils diffèrent par la façon dont ils permettent l'interaction à l'échelle de la scène, ce qui se réfère à l'échange d'informations entre les jetons à travers une scène. L'attention de fenêtre réduit le coût de calcul en restreignant l'attention à des groupes de jetons appelés fenêtres; dans les conceptions sérialisées, chaque fenêtre contient un nombre fixe de jetons consécutifs d'une séquence ordonnée spatialement. Parce que chaque bloc ne connecte directement que les jetons au sein de la même fenêtre, l'échange d'informations à travers les fenêtres repose sur les changements d'adhésion de fenêtre à travers

Machine translation (Meta M2M100); the English original follows.

Abstract

Point cloud transformers aim to capture detailed local geometry and relationships across a scene at a manageable computational cost. Two common approaches are window attention and global attention. They differ in how they enable scene-wide interaction, which refers to information exchange among tokens across a scene. Window attention reduces computational cost by restricting self-attention to groups of tokens called windows; in serialized designs, each window contains a fixed number of consecutive tokens from a spatially ordered sequence. Because each block only directly connects tokens within the same window, information exchange across windows relies on changes in window membership across blocks and hierarchical pooling. Global attention, in contrast, directly connects all tokens within a scene, but computing attention over all token pairs can be costly when the number of tokens is large. To balance computational cost and scene-wide interaction, we introduce MixPT, a point transformer that uses a mix of window and global attention blocks in an interleaved schedule, while retaining a single fixed serialization pattern for its window blocks. For scenes with a large number of tokens, MixPT applies global attention to spatially pooled tokens and returns residual updates to the original-resolution stream. Across indoor and outdoor benchmarks, MixPT achieves competitive accuracy with lower latency than representative window and global attention baselines. We further demonstrate the feasibility of scaling the attention stage in depth by repeating the interleaved pattern of window and global attention blocks.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Window (computing), Serialization, Security token, Transformer, Point cloud

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange