English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Transformers have been recently explored for 3D point cloud understandingwith impressive progress achieved. A large number of points, over 0.1 million,make the global self-attention infeasible for point cloud data. Thus, mostmethods propose to apply the transformer in a local region, e.g., spherical orcubic window. However, it still contains a large number of Query-Key pairs,which requires high computational costs. In addition, previous methods usuallylearn the query, key, and value using a linear projection without modeling thelocal 3D geometric structure. In this paper, we attempt to reduce the costs andmodel the local geometry prior by developing a new transformer block, namedConDaFormer. Technically, ConDaFormer disassembles the cubic window into threeorthogonal 2D planes, leading to fewer points when modeling the attention in asimilar range. The disassembling operation is beneficial to enlarging the rangeof attention without increasing the computational complexity, but ignores somecontexts. To provide a remedy, we develop a local structure enhancementstrategy that introduces a depth-wise convolution before and after theattention. This scheme can also capture the local geometric information. Takingadvantage of these designs, ConDaFormer captures both long-range contextualinformation and local priors. The effectiveness is demonstrated by experimentalresults on several 3D point cloud understanding benchmarks. Code is availableat https://github.com/LHDuan/ConDaFormer .

ConDaFormer: Disassembled Transformer with Local Structure Enhancement for 3D Point Cloud Understanding