用Vulkan实现高性能3D高斯溅射训练，跨显卡兼容且提速三倍。
VkSplat: High-Performance 3DGS Training in Vulkan Compute
- 全用Vulkan计算实现3D高斯溅射训练，避开CUDA依赖。
- 相比CUDA+PyTorch基线提速3.3倍，显存减少33%。
- 适合需要跨平台部署的3D重建与图形研究者。
我们提出VkSplat，一个完全基于Vulkan计算的高性能、跨厂商3D高斯溅射（3DGS）训练管道，解决了现有训练方案在性能和兼容性上的局限。通过多种优化，我们在保持质量的前提下，相较CUDA+PyTorch基线实现了3.3倍的加速和33%的显存降低，并验证了在不同GPU厂商间的兼容性。据我们所知，这是首个达到顶尖性能的全Vulkan 3DGS训练管道。代码开源：https://github.com/harry7557558/vksplat。
原文摘要 · Abstract (English)
We present VkSplat, a high-performance, cross-vendor 3D Gaussian Splatting (3DGS) training pipeline implemented fully in Vulkan compute, addressing performance and compatibility limitation of existing training pipelines. With various optimizations, we achieve $3.3\times$ speed and $33\%$ VRAM reduction over CUDA+PyTorch baseline, maintaining quality, and demonstrating compatibility across GPU vendors. To the best of our knowledge, this is the first fully-Vulkan-based 3DGS training pipeline that achieves state-of-the-art performance. Code: \href{https://github.com/harry7557558/vksplat}{https://github.com/harry7557558/vksplat}
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。