Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

VkSplat: High-Performance 3DGS Training in Vulkan Compute

About

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}

Jingxiang Chen, Mohamed Ibrahim, Yang Liu• 2026

Related benchmarks

TaskDatasetResultRank
Novel View SynthesisMip-NeRF 360
PSNR29.2
44
3D Gaussian Splatting TrainingMip-NeRF 360 Bicycle scene (train)
Total Time (s)285
4
Showing 2 of 2 rows

Other info

Follow for update