LitePT：更轻量却更强的点云 Transformer
LitePT: Lighter Yet Stronger Point Transformer
为 3D 点云骨干网络提供了「浅层卷积 + 深层注意力」的分层设计准则，并给出无需额外参数的 PointROPE 三维位置编码，可直接替换 Point Transformer 系列骨干以降低显存与延迟。
原文依据 we propose a new, improved 3D point cloud backbone that employs convolutions in early stages and switches to attention for deeper layers
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论面向三维点云处理的现代神经网络架构同时包含卷积层与注意力模块，但如何将它们组合起来仍不明确。我们分析了三维点云网络中不同计算模块的作用，发现了一种符合直觉的行为：在高分辨率的前期层中，卷积足以提取低层几何信息，此时注意力代价高昂且不带来任何收益；而在低分辨率的深层中，注意力能更高效地捕捉高层语义与上下文。基于这一设计原则，我们提出了一种新的、改进的三维点云骨干网络，在前期阶段使用卷积，并在更深的层切换为注意力。为避免在丢弃冗余卷积层时损失空间布局信息，我们引入了一种新颖的、无参数的三维位置编码 PointROPE。由此得到的 LitePT 模型相比最优的 Point Transformer V3 参数量减少 3.6 倍、运行速度快 2 倍、内存占用减少 2 倍，却在一系列任务与数据集上达到与之相当甚至更优的性能。
英文原文摘要 1084 字符
Modern neural architectures for 3D point cloud processing contain both convolutional layers and attention blocks, but the best way to assemble them remains unclear. We analyse the role of different computational blocks in 3D point cloud networks and find an intuitive behaviour: convolution is adequate to extract low-level geometry at high-resolution in early layers, where attention is expensive without bringing any benefits; attention captures high-level semantics and context in low-resolution, deep layers more efficiently. Guided by this design principle, we propose a new, improved 3D point cloud backbone that employs convolutions in early stages and switches to attention for deeper layers. To avoid the loss of spatial layout information when discarding redundant convolution layers, we introduce a novel, parameter-free 3D positional encoding, PointROPE. The resulting LitePT model has 3.6x fewer parameters, runs 2x faster, and uses 2x less memory than the state-of-the-art Point Transformer V3, but nonetheless matches or outperforms it on a range of tasks and datasets.