The quickly growing open source platform LichtFeld Studio has just released V0.5.2. Earlier this year, the v0.5 release brought embedded Python runtime and its plugin marketplace. V0.5.2 has added several smaller updates to better leverage the foundation that has been laid.
The feature that will get the most immediate use is probably HDR image map support for rendering. Accurate environment lighting has been a persistent gap in 3DGS viewers and training tools alike, and being able to bring in real HDR maps for rendering gives artists a much more faithful sense of how a scene will look under intended lighting conditions. The background color picker is a related addition.
More analytically interesting is the visualization of loss per camera. During training, understanding which cameras are contributing poorly, and why, has historically required digging into logs or writing custom tooling. Surfacing this directly in the UI makes it considerably easier to identify problematic capture angles, noisy input frames, or geometry that isn’t reconstructing well, without leaving the application. I am hoping to better leverage this information into SplatKing. Training evaluation more broadly now works from within the UI as well, which closes another loop that previously required dropping to the command line.
On the interface side, a dockable sequencer and histogram, along with the free orbit and FPV camera modes have been added. The histogram in SuperSplat is one of the most underrated tools and so I am excited to see this make its way into LFS.
Underneath all of this, the migration from ImGui to RmlUi continues. It’s infrastructure work that won’t be visible to most users in v0.5.2, but RmlUi’s HTML and CSS like approach to layout and styling means the interface can be iterated on much more rapidly going forward, and the visual consistency it enables will compound across releases.
The release also brings a development on the sustainability side. Donation support through portal.lichtfeld.io has now enabled the project to issue its first contributor grant to Shady Gmira, who will be working on the upcoming v0.5.3 release. LichtFeld Studio’s long term goal of expanding these grants to more active contributors is starting to look credible, backed by foundational sponsor Core11 GmbH and gold sponsor Volinga.
V0.5.3 has a clear direction to move to Vulkan for the frontend, which opens the door to vendor independent GPU support and enables initial path tracing and custom shaders. Direct support for the sparkjs RAD format and asset management for training data and 3DGS files are also on the list. LichtFeld Studio remains free and open source to build from source. If you would like precompiled binaries, MrNeRF is asking for a $30 donation to help fund the project.
More information can be found here.
New to Gaussian Splatting? Start here
