XVerse, which offers a free 3DGS plugin for Unreal Engine 5.3, has announced the launch of their local training toolkit, XV3DTools. This toolkit enables one-click training of 3DGS models from videos for use in Unreal Engine 5.
The requirements for downloading are minimal, although it is currently only compatible with Windows and requires at least CUDA 11.
The most significant feature this time is the ability to train videos locally into 3DGS within Unreal Engine. They specify that it inputs .mp4s to train. While it may not be the fastest trainer, they are currently training on a 2080. As you might imagine, XV3DTools is completely compatible with their plugin and similarly carries an Apache 2.0 license and can be used in UE 5.3.
While they specified that the training platform works with video, I imagine that you would be able to use ffmpeg to stitch a collection of images into a video and then just align the total number of frames used with the video. Their plugin can be downloaded from their Releases on Github here and XV3DTools can be downloaded here.
Speaking of the XV3DGS-UEPlugin, they have also released an updated version v1.1, which enhances the foundational rendering features of 3DGS in UE. New features include automatic LOD generation, supporting over 200,000 point clouds in Niagara.
New to Gaussian Splatting? Start here
