The point cloud learning community witnesses a modeling shift from CNNs to Transformers, where pure Transformer architectures have achieved top accuracy on the major learning benchmarks. However, existing point Transformers are computationally expensive since they need to generate a large attention map, which has quadratic complexity (both in space and time) with respect to input size. To solve this shortcoming, we introduce Patch ATtention (PAT) to adaptively learn a much smaller set of bases upon which the attention maps are computed. By a weighted summation upon these bases, PAT not only captures the global shape context but also achieves linear complexity to input size. In addition, we propose a lightweight Multi-Scale aTtention (MST) block to build attentions among features of different scales, providing the model with multi-scale features. Equipped with the PAT and MST, we construct our neural architecture called PatchFormer that integrates both modules into a joint framework for point cloud learning. Extensive experiments demonstrate that our network achieves comparable accuracy on general point cloud learning tasks with$9.2\times$speed-up than previous point Transformers.

The recently developed pure transformer architectures have attained promising accuracy on point cloud learning benchmarks compared to convolutional neural networks. However, existing point cloud Transformers are computationally expensive because they waste a significant amount of time on structuring irregular data. To solve this shortcoming, we present the Sparse Window Attention module to gather coarse-grained local features from nonempty voxels. The module not only bypasses the expensive irregular data structuring and invalid empty voxel computation, but also obtains linear computational complexity with respect to voxel resolution. Meanwhile, we leverage two different self-attention variants to gather fine-grained features about the global shape according to different scale of point clouds. Finally, we construct our neural architecture called point-voxel transformer (PVT), which integrates these modules into a joint framework for point cloud learning. Compared with previous transformer-based and attention-based models, our method attains a top accuracy of 94.1% on the classification benchmark and 10 × $10\times $ inference speedup on average. Extensive experiments also validate the effectiveness of PVT on semantic segmentation benchmarks. Our code and pretrained model are avaliable at https://github.com/HaochengWan/PVT.