harry7557558/spirula-studioPublic

Cross-vendor 3D Gaussian Splatting trainer - video to splat to mesh, Vulkan or CUDA.

AI summary: A standalone, cross-vendor 3D Gaussian Splatting trainer that converts videos to meshes via Vulkan or CUDA.

Stars
2.5K
+87 today
Forks
243
Watchers
40
Open issues
88
Open PRs
23
Contributors
~33
Commits
1.2K
Branches
25

C++GPL-3.0Created May 4, 2024Last push 2d agoLatest release v2026.11.12+184 stars this week+722 this month

Quick answers

What is spirula-studio?
A standalone, cross-vendor 3D Gaussian Splatting trainer that converts videos to meshes via Vulkan or CUDA.
What does spirula-studio do?
Spirula Studio is a self-contained application for training 3D Gaussian Splatting models directly from raw photos or videos. It manages the entire pipeline, from generating splats to producing textured meshes, without requiring complex Python, PyTorch, or COLMAP installations. The software is highly optimized, capable of training massive models in just 8 GB of VRAM. It uniquely supports cross-vendor hardware, running seamlessly on NVIDIA, AMD, Intel, and Apple GPUs via the Vulkan API. This dramatically lowers the barrier to entry for high-quality 3D reconstruction.
Who is spirula-studio for?
3D artists, game developers, and hobbyists wanting to experiment with Gaussian Splatting without complex setups. It is ideal for users who do not have high-end NVIDIA hardware but still want fast 3D reconstruction.
How popular is spirula-studio on GitHub?
harry7557558/spirula-studio has 2,508 stars and 243 forks on GitHub, and gained 184 stars in the last 7 days.
What license does spirula-studio use?
harry7557558/spirula-studio is released under the GPL-3.0 license.

Star history

since Sep 23, 2026
01K2KSep 2026Oct 2026Oct 2026Nov 2026
2.5K stars as of Nov 19, 2026. Measured daily since Sep 23, 2026; GitHub no longer exposes earlier star timestamps.

Update history

4 recorded
  • Nov 19, 2026New release v2026.11.12 (previously v2026.10.30).
  • Nov 7, 2026New release v2026.10.30 (previously v2026.10.16).
  • Oct 20, 2026New release v2026.10.16 (previously v2026.10.6).
  • Oct 11, 2026New release v2026.10.6 (previously v2026.9.30).

Contribution activity

commits per day, last 52 weeks
NovDecJanFebMarAprMayJunJulAugSepOctNovMonWedFri2025-11-23: 2 commits2025-11-24: 0 commits2025-11-25: 1 commit2025-11-26: 2 commits2025-11-27: 1 commit2025-11-28: 1 commit2025-11-29: 1 commit2025-11-30: 1 commit2025-12-01: 0 commits2025-12-02: 2 commits2025-12-03: 0 commits2025-12-04: 0 commits2025-12-05: 0 commits2025-12-06: 2 commits2025-12-07: 0 commits2025-12-08: 0 commits2025-12-09: 0 commits2025-12-10: 0 commits2025-12-11: 0 commits2025-12-12: 0 commits2025-12-13: 0 commits2025-12-14: 1 commit2025-12-15: 1 commit2025-12-16: 0 commits2025-12-17: 0 commits2025-12-18: 1 commit2025-12-19: 0 commits2025-12-20: 0 commits2025-12-21: 0 commits2025-12-22: 0 commits2025-12-23: 0 commits2025-12-24: 0 commits2025-12-25: 0 commits2025-12-26: 0 commits2025-12-27: 2 commits2025-12-28: 1 commit2025-12-29: 0 commits2025-12-30: 1 commit2025-12-31: 1 commit2026-01-01: 3 commits2026-01-02: 0 commits2026-01-03: 0 commits2026-01-04: 0 commits2026-01-05: 1 commit2026-01-06: 0 commits2026-01-07: 0 commits2026-01-08: 0 commits2026-01-09: 0 commits2026-01-10: 1 commit2026-01-11: 2 commits2026-01-12: 0 commits2026-01-13: 0 commits2026-01-14: 0 commits2026-01-15: 0 commits2026-01-16: 0 commits2026-01-17: 1 commit2026-01-18: 0 commits2026-01-19: 1 commit2026-01-20: 0 commits2026-01-21: 0 commits2026-01-22: 0 commits2026-01-23: 0 commits2026-01-24: 1 commit2026-01-25: 0 commits2026-01-26: 1 commit2026-01-27: 1 commit2026-01-28: 2 commits2026-01-29: 0 commits2026-01-30: 0 commits2026-01-31: 0 commits2026-02-01: 4 commits2026-02-02: 1 commit2026-02-03: 3 commits2026-02-04: 0 commits2026-02-05: 1 commit2026-02-06: 3 commits2026-02-07: 2 commits2026-02-08: 1 commit2026-02-09: 3 commits2026-02-10: 1 commit2026-02-11: 0 commits2026-02-12: 2 commits2026-02-13: 0 commits2026-02-14: 3 commits2026-02-15: 6 commits2026-02-16: 2 commits2026-02-17: 2 commits2026-02-18: 2 commits2026-02-19: 2 commits2026-02-20: 0 commits2026-02-21: 2 commits2026-02-22: 2 commits2026-02-23: 1 commit2026-02-24: 3 commits2026-02-25: 1 commit2026-02-26: 0 commits2026-02-27: 0 commits2026-02-28: 1 commit2026-03-01: 4 commits2026-03-02: 2 commits2026-03-03: 1 commit2026-03-04: 2 commits2026-03-05: 0 commits2026-03-06: 1 commit2026-03-07: 3 commits2026-03-08: 1 commit2026-03-09: 1 commit2026-03-10: 3 commits2026-03-11: 1 commit2026-03-12: 1 commit2026-03-13: 1 commit2026-03-14: 5 commits2026-03-15: 2 commits2026-03-16: 2 commits2026-03-17: 1 commit2026-03-18: 2 commits2026-03-19: 3 commits2026-03-20: 2 commits2026-03-21: 2 commits2026-03-22: 5 commits2026-03-23: 2 commits2026-03-24: 1 commit2026-03-25: 1 commit2026-03-26: 0 commits2026-03-27: 0 commits2026-03-28: 1 commit2026-03-29: 1 commit2026-03-30: 0 commits2026-03-31: 2 commits2026-04-01: 0 commits2026-04-02: 0 commits2026-04-03: 1 commit2026-04-04: 0 commits2026-04-05: 0 commits2026-04-06: 0 commits2026-04-07: 0 commits2026-04-08: 0 commits2026-04-09: 1 commit2026-04-10: 2 commits2026-04-11: 3 commits2026-04-12: 0 commits2026-04-13: 2 commits2026-04-14: 1 commit2026-04-15: 3 commits2026-04-16: 0 commits2026-04-17: 3 commits2026-04-18: 2 commits2026-04-19: 5 commits2026-04-20: 3 commits2026-04-21: 3 commits2026-04-22: 1 commit2026-04-23: 2 commits2026-04-24: 2 commits2026-04-25: 4 commits2026-04-26: 0 commits2026-04-27: 2 commits2026-04-28: 0 commits2026-04-29: 4 commits2026-04-30: 2 commits2026-05-01: 0 commits2026-05-02: 2 commits2026-05-03: 1 commit2026-05-04: 2 commits2026-05-05: 1 commit2026-05-06: 3 commits2026-05-07: 3 commits2026-05-08: 0 commits2026-05-09: 1 commit2026-05-10: 1 commit2026-05-11: 0 commits2026-05-12: 1 commit2026-05-13: 1 commit2026-05-14: 2 commits2026-05-15: 2 commits2026-05-16: 5 commits2026-05-17: 2 commits2026-05-18: 2 commits2026-05-19: 2 commits2026-05-20: 5 commits2026-05-21: 1 commit2026-05-22: 3 commits2026-05-23: 2 commits2026-05-24: 2 commits2026-05-25: 1 commit2026-05-26: 5 commits2026-05-27: 3 commits2026-05-28: 8 commits2026-05-29: 6 commits2026-05-30: 4 commits2026-05-31: 4 commits2026-06-01: 1 commit2026-06-02: 4 commits2026-06-03: 2 commits2026-06-04: 5 commits2026-06-05: 3 commits2026-06-06: 5 commits2026-06-07: 9 commits2026-06-08: 4 commits2026-06-09: 5 commits2026-06-10: 2 commits2026-06-11: 0 commits2026-06-12: 2 commits2026-06-13: 3 commits2026-06-14: 2 commits2026-06-15: 5 commits2026-06-16: 2 commits2026-06-17: 3 commits2026-06-18: 0 commits2026-06-19: 5 commits2026-06-20: 1 commit2026-06-21: 4 commits2026-06-22: 1 commit2026-06-23: 2 commits2026-06-24: 2 commits2026-06-25: 2 commits2026-06-26: 1 commit2026-06-27: 0 commits2026-06-28: 0 commits2026-06-29: 0 commits2026-06-30: 0 commits2026-07-01: 2 commits2026-07-02: 2 commits2026-07-03: 2 commits2026-07-04: 2 commits2026-07-05: 3 commits2026-07-06: 3 commits2026-07-07: 1 commit2026-07-08: 2 commits2026-07-09: 0 commits2026-07-10: 6 commits2026-07-11: 6 commits2026-07-12: 7 commits2026-07-13: 0 commits2026-07-14: 1 commit2026-07-15: 2 commits2026-07-16: 3 commits2026-07-17: 6 commits2026-07-18: 4 commits2026-07-19: 6 commits2026-07-20: 6 commits2026-07-21: 4 commits2026-07-22: 5 commits2026-07-23: 0 commits2026-07-24: 0 commits2026-07-25: 0 commits2026-07-26: 0 commits2026-07-27: 3 commits2026-07-28: 4 commits2026-07-29: 2 commits2026-07-30: 1 commit2026-07-31: 1 commit2026-08-01: 5 commits2026-08-02: 0 commits2026-08-03: 6 commits2026-08-04: 2 commits2026-08-05: 4 commits2026-08-06: 3 commits2026-08-07: 4 commits2026-08-08: 7 commits2026-08-09: 4 commits2026-08-10: 1 commit2026-08-11: 3 commits2026-08-12: 3 commits2026-08-13: 6 commits2026-08-14: 7 commits2026-08-15: 3 commits2026-08-16: 7 commits2026-08-17: 7 commits2026-08-18: 5 commits2026-08-19: 3 commits2026-08-20: 6 commits2026-08-21: 2 commits2026-08-22: 9 commits2026-08-23: 4 commits2026-08-24: 2 commits2026-08-25: 4 commits2026-08-26: 4 commits2026-08-27: 4 commits2026-08-28: 0 commits2026-08-29: 6 commits2026-08-30: 5 commits2026-08-31: 8 commits2026-09-01: 13 commits2026-09-02: 3 commits2026-09-03: 3 commits2026-09-04: 7 commits2026-09-05: 1 commit2026-09-06: 4 commits2026-09-07: 9 commits2026-09-08: 4 commits2026-09-09: 5 commits2026-09-10: 3 commits2026-09-11: 4 commits2026-09-12: 6 commits2026-09-13: 1 commit2026-09-14: 2 commits2026-09-15: 2 commits2026-09-16: 2 commits2026-09-17: 5 commits2026-09-18: 4 commits2026-09-19: 7 commits2026-09-20: 5 commits2026-09-21: 7 commits2026-09-22: 6 commits2026-09-23: 11 commits2026-09-24: 3 commits2026-09-25: 4 commits2026-09-26: 3 commits2026-09-27: 5 commits2026-09-28: 6 commits2026-09-29: 8 commits2026-09-30: 11 commits2026-10-01: 5 commits2026-10-02: 5 commits2026-10-03: 2 commits2026-10-04: 3 commits2026-10-05: 7 commits2026-10-06: 3 commits2026-10-07: 3 commits2026-10-08: 10 commits2026-10-09: 1 commit2026-10-10: 7 commits2026-10-11: 9 commits2026-10-12: 4 commits2026-10-13: 2 commits2026-10-14: 2 commits2026-10-15: 1 commit2026-10-16: 0 commits2026-10-17: 2 commits2026-10-18: 8 commits2026-10-19: 5 commits2026-10-20: 4 commits2026-10-21: 0 commits2026-10-22: 1 commit2026-10-23: 1 commit2026-10-24: 0 commits2026-10-25: 4 commits2026-10-26: 3 commits2026-10-27: 8 commits2026-10-28: 3 commits2026-10-29: 13 commits2026-10-30: 3 commits2026-10-31: 1 commit2026-11-01: 3 commits2026-11-02: 0 commits2026-11-03: 0 commits2026-11-04: 1 commit2026-11-05: 9 commits2026-11-06: 8 commits2026-11-07: 7 commits2026-11-08: 9 commits2026-11-09: 0 commits2026-11-10: 6 commits2026-11-11: 10 commits2026-11-12: 10 commits2026-11-13: 1 commit2026-11-14: 4 commits2026-11-15: 6 commits2026-11-16: 2 commits2026-11-17: 8 commits2026-11-18: 9 commits2026-11-19: 0 commits2026-11-20: 0 commits2026-11-21: 0 commits
917 commits in the last yearLessMore

Signals and awards

derived from tracked data
  • Rising fast

    +184 stars this week

  • Very active

    917 commits in 52 weeks

  • Continuous integration

    Automated checks passing

What spirula-studio does

Spirula Studio is a self-contained application for training 3D Gaussian Splatting models directly from raw photos or videos. It manages the entire pipeline, from generating splats to producing textured meshes, without requiring complex Python, PyTorch, or COLMAP installations. The software is highly optimized, capable of training massive models in just 8 GB of VRAM. It uniquely supports cross-vendor hardware, running seamlessly on NVIDIA, AMD, Intel, and Apple GPUs via the Vulkan API. This dramatically lowers the barrier to entry for high-quality 3D reconstruction.

3D artists, game developers, and hobbyists wanting to experiment with Gaussian Splatting without complex setups. It is ideal for users who do not have high-end NVIDIA hardware but still want fast 3D reconstruction.

  • End-to-end pipeline: Processes raw video or images into 3D splats and textured meshes in one tool.
  • Cross-vendor GPU support: Utilizes Vulkan to run on AMD, Intel, and Apple hardware, alongside CUDA for NVIDIA.
  • Self-contained binary: Eliminates dependency hell by removing the need for Python, PyTorch, or COLMAP.
  • VRAM optimization: Trains up to 10 million full-spherical harmonics Gaussians using only 8 GB of video memory.
  • Quantized training: Employs memory-efficient techniques to accelerate the 3D reconstruction process.

Where teams use it

Accessible 3D reconstruction

Generate high-quality 3D models from smartphone video without needing a deep learning engineering background.

Non-NVIDIA rendering

Train Gaussian Splatting models natively on MacBooks or AMD-equipped PCs using the Vulkan backend.

Low-VRAM environments

Process large scenes on consumer-grade GPUs by leveraging the tool's aggressive memory optimizations.

Mesh generation

Convert trained Gaussian splats directly into standard textured meshes for use in game engines or traditional 3D software.

README

master branch

Spirula Studio

GPLv3 License  GitHub Releases  Platform

Download • Build from Source • Gallery • Web Viewer

Spirula Studio trains 3D Gaussian Splatting models – from raw photo/video to splat to textured mesh – in one self-contained binary. No Python/PyTorch, no separate COLMAP install. Runs on NVIDIA, AMD, Intel, and Apple GPUs via Vulkan, trains 10M full-SH Gaussians in 8 GB VRAM, and has native support for fisheye and 360° cameras.

Spirula Studio - Open Source 3D Gaussian Splatting Pipeline

Dataset credit: Garage by Simon Bethke (CC BY-SA 4.0); Flight Systems and Control Lab at UTIAS; MegaDepth-X; Mip-NeRF 360.

Features

  • Cross vendor support via Vulkan compute – Runs on NVIDIA, AMD, Intel, and Apple GPUs

  • One strategy combining advantages of MCMC/IGS+/MRNF – Sharper results, fewer floaters, from objects to large scenes

  • Extreme VRAM efficiency with quantized training – Up to 10 million SH3 Gaussians in 8GB VRAM

  • Native 360° camera and equirectangular support – Load a dataset and train, no undistortion needed

  • Modified Bilateral grid and PPISP for exposure/WB correction – Improving quality without unwanted color shift or darkening

  • Built-in lightning-fast SfM, AI masking, frame extraction from videos – No need to wait for COLMAP or run separate scripts

  • Depth/normal, meshing, skybox, linear color... And more.

News

  • September 23, 2026: Editing and rendering features – Editing features for 3DGS models, sparse reconstructions, meshes, and masks have been added, along with support for exporting image and video renders.

  • September 10, 2026: Metric scale – The dataset creation module now uses telemetry metadata in common video and image formats to recover metric scale and orientation, addressing the popular report that reconstruction results are too large/small or tilted.

  • September 3, 2026: LoMa feature support – The SfM module now supports LoMa for feature detection and matching on difficult datasets.

  • August 14, 2026: macOS support – Support for training on macOS/Apple Silicon has been validated. The app can now be downloaded from Releases page.

  • August 8, 2026: Multilingual support – Multilingual support has been added, available to both GUI and CLI. Supported languages: English, 日本語, 简体中文, 繁體中文, 한국어, Deutsch, Français, Español, Português, Italiano, Nederlands, Русский, Türkçe.

Download

Binaries for Windows, Linux, and macOS can be downloaded from Releases page. Simply select the one for your platform, download and unzip, and double click to open the GUI.

If you are training on remote/cloud GPUs, you may use the CLI – Run spirula --help for details. By default, spirula train command will serve a viewer on an HTTP port, one you can forward over ssh and view training progress in your web browser.

Build from source

To build from source, Spirula Studio provides two backends:

  • Vulkan (Recommended): The cross-platform and cross-vendor option. Most tested. Works on all major GPUs. Faster to build and produces smaller binary.

  • CUDA: Legacy option for CUDA-capable NVIDIA GPUs.

Both provide the same training and meshing functionality. CUDA backend may be faster or slower than Vulkan depending on GPU driver, with difference generally within a few percents. Vulkan backend can be slightly more VRAM efficient in some cases.

Backend GPU/Vendor Support Platform Support Dependencies Additional Features
Vulkan NVIDIA, AMD, Intel, Apple Silicon Windows, Linux, macOS Vulkan/MoltenVK, CMake/Ninja Native support for SfM, frame extraction from videos, and AI masking
CUDA Most NVIDIA GPUs Windows, Linux CUDA, CMake/Ninja -
Details for building the Vulkan backend

Make sure you have Vulkan SDK installed. On macOS, MoltenVK is automatically fetched by CMake. Clone the repository and run the commands:

Windows with MSVC:

cd spirula-studio\
.\build_develop.bat -DSS_BACKEND=vulkan -DSS_ENABLE_PATENTED=ON

If it builds successfully, you get build_vulkan\spirula.exe.

Windows with GCC/Clang:

cd spirula-studio\
cmake -G Ninja -B build_vulkan -DCMAKE_BUILD_TYPE=Release -DSS_BACKEND=vulkan -DSS_ENABLE_PATENTED=ON -DCMAKE_MAKE_PROGRAM=Ninja
cmake --build build_vulkan -j

Pass -DCMAKE_C_COMPILER and -DCMAKE_CXX_COMPILER to the first cmake command if needed.

If it builds successfully, you get build_vulkan\spirula.exe.

Linux:

cd spirula-studio/
bash build_develop.bash -DSS_BACKEND=vulkan -DSS_ENABLE_PATENTED=ON

If it builds successfully, you get build_vulkan/spirula binary.

macOS:

cd spirula-studio/
bash build_develop.bash -DSS_BACKEND=vulkan -DSS_ENABLE_PATENTED=ON
cmake --build build --target macos_app
cmake --build build --target macos_dmg

macOS has only the one backend, so it builds into build/ rather than into a per-backend tree. If it builds successfully, you get build/spirula binary similar to Linux. Additionally, it wraps that binary in a double-clickable build/Spirula Studio.app, as well as disk image build/Spirula Studio.dmg. MoltenVK is statically linked by default and will run on a Mac without dependency installed.

Notes regarding third-party licensing

-DSS_ENABLE_PATENTED=ON enables decoding video on the GPU instead of shelling out to ffmpeg (about 15x faster frame extraction, and without need to install ffmpeg). However, AVC/HEVC bitstream parsers carry third-party patent exposure. If you turn this on, you are responsible for ensuring compliance with local patent laws regarding AVC/HEVC playback.

Masking needs a SAM checkpoint, which the GUI downloads on first use and caches. The checkpoints are Meta's models under Meta's licenses – SAM 2.1 is Apache-2.0, SAM 3 is under Meta's own, non-standard license. They are never bundled, and the GUI shows the terms before fetching anything. On the command line, point --model at a file you downloaded yourself.

Details for building the CUDA backend

Make sure you have a recent version of CUDA installed. On Windows, you also need MSVC compiler compatible with your CUDA version. Clone the repository and run the commands:

Windows:

cd spirula-studio\
.\build_develop.bat -DSS_BACKEND=cuda

If it builds successfully, you get build_cuda\spirula.exe.

Linux:

cd spirula-studio/
bash build_develop.bash -DSS_BACKEND=cuda

If it builds successfully, you get build_cuda/spirula binary.

Gallery

You can find some professional-quality splats trained by Spirula Studio from Megascapes Library and their SuperSplat page.

Collection of splats created by the users of Spirula Studio can also be found on SuperSplat page.

Some splats created by the author of Spirula Studio can also be found on my SuperSplat page.

Trivia

Spirula Studio (formerly spirulae-splat) is named after the now-inactive project spirulae, which was named after the deep-ocean cephalopod mollusk.

Spirula Studio is maintained almost entirely by one person. Issues and PRs welcome – I sometimes respond late, but rest assured that I do review them all.

View on GitHub

Recent activity

commits and pull requests

Recent open issues

view all

Releases and announcements

17 total
  1. v2026.11.12v2026.11.12Nov 12, 20264.5K downloads

    Changelog - Improved training for better distant details - Support correcting camera calibration residual (like rolling shutter) during training - Improved floater suppression in training - Speed up masking and normal/depth estimation on Apple Silicon - Shortened tooltip and other help text to make UI easier to read - Added GLOMAP-style global mapper for SfM - Added FlexGlue feature matcher for SfM - Better support for saved fixed-area masks - Added blending option to better visualize camera calibration error in training GUI - Fixed instability with HDR training - Optimized native PLY viewer for viewing large 3DGS PLYs, like those merged from partitioned training - Added option to automatically start training after dataset creation - Added option to run a command after training - Support visualizing camera exposure information Notes - This release carries training quality difference intended to be improvement. Speed and VRAM overhead is generally within a few percent. Please submit an issue if you notice significant quality or efficiency regression. - In training GUI, floater suppression levels have been changed from mild/strong to mild/moderate/strong, with the

  2. v2026.10.30v2026.10.30Oct 30, 20267.4K downloads

    Changelog - Improved training quality in general - Improved compatibility of trained splats with PlayCanvas and Spark viewers - Fixed incompatibility with older AMD GPUs - Fixed occasional training out of memory caused by VRAM spike - Improved accuracy of progress bars, and added RAM/VRAM forecasting for SfM - Support MoGe 3 and Marigold v2 for depth/normal estimation - Support scaling the model based on distance between two points in editor - Support recovering orientation and scale based on Apriltags and ArUco markers in editor (experimental) - Support skipping blurry and near-duplicate frames during dataset creation - Support manually removing false matches in dataset creation - Support parsing masks with file names in Reality Scan convention - Support direct camera control beside preview pane during trajectory editing - Support reading GPS origin from gauge.txt - Support dataset creation from QooCam 3 Ultra raw footage Notes - This release carries training quality difference intended to be improvement. Please submit an issue if you notice significant regression in training quality or efficiency on your datasets. - A lot of great new features in this release a

  3. v2026.10.16v2026.10.16Oct 16, 20266.3K downloads

    Changelog - Optimized RAM and VRAM usage for SfM on large datasets - Optimized VRAM usage when training on high resolution images - Support dense initialization, inspired by [LichtFeld Densification Plugin](https://github.com/shadygm/lichtfeld-densification-plugin) - Support taking screenshots and saving camera views during training and viewing - Better support for creating datasets on log footage (DJI D-Log M, Insta360 I-Log, etc.) - Improved use of GPS in dataset creation - Support source preview and video trimming in dataset creation - Support spherical beta from [Deformable Beta Splatting](https://rongliu-leo.github.io/beta-splatting/) for view-dependent appearance - Reduced chance of watchdog timeout error (VkResult -4 / "Impacting Interactivity") on slower GPUs - Reduced seams in merged models after partitioned training - Revised masking prompts for dataset creation with 360 cameras - Made scene centering automatic for datasets in ECEF-like frames to reduce precision issues - Support parsing IMU and GPS from Antigravity A1 footage Notes - This release carries RAM and VRAM optimizations in SfM and training. Expect similar training speed, SfM speed can be simil

  4. v2026.10.6v2026.10.6Oct 6, 20264.4K downloads

    Changelog - Support region of interest (ROI) in training - Redesigned GUI home screen to keep recently opened items of various formats - Faster depth and normal map estimation for 360 cameras - Support separate masks for feature points and training (useful for sky and texture-rich clouds) - Support using LiDAR scans to recover metric scale and compute depth and normal maps - Reduced unnecessary step reruns when updating datasets after adjusting parameters - Improved support for various input image formats - Improved support for utilizing GPS in dataset creation - Addressed occasional training divergence - Addressed occasional crashes during meshing - Added warning message on known incompatible hardware and drivers - Support zooming using mouse wheel when editing masks Notes: - This release carries difference in both training and dataset creation intended to be improvement. Please submit an issue if you notice significant regression in quality or resource usage compared to previous versions. - LiDAR support was tested on a limited set of publicly available datasets. It is currently an experimental feature. Community feedback is welcome.

  5. v2026.9.30v2026.9.30Sep 30, 20262.7K downloads

    Changelog - General speed improvement in dataset creation - Support splitting a large dataset into multiple parts, then train separately and merge - Support natural language prompt for SAM 2.1 masking models - Support [BiRefNet](https://github.com/zhengpeng7/birefnet) masking model - Improved training ETA estimation, and added training VRAM forecasting and OOM warning - Reduced chance of "VkResult -4" error in dataset creation and meshing - Improved support for using telemetry information in dataset creation - Support pen tool for masks - Support using a separate fixed-area mask for each camera in dual-fisheye setups - Fixed error loading PLY saved from a resumed training - Fixed COLMAP dataset parsing error with relative paths - Fixed video decoding issues with some drivers Notes: - This release carries difference in SfM intended to be improvement. Please submit an issue if you notice significant regression in accuracy or speed compared to previous versions. - The default masking model has been changed from SAM 3 to SAM 2.1 Base+ with Grounding DINO, which provides 2-3x faster masking with slightly reduced accuracy. SAM 3 remains available as a non-default option.

Code frequency

additions and deletions
+311K-311KWeek of 2025-11-23: +2,247 linesWeek of 2025-11-23: -1,793 linesWeek of 2025-11-30: +11,983 linesWeek of 2025-11-30: -5,657 linesWeek of 2025-12-07: +0 linesWeek of 2025-12-07: -0 linesWeek of 2025-12-14: +17,577 linesWeek of 2025-12-14: -17,210 linesWeek of 2025-12-21: +622 linesWeek of 2025-12-21: -61 linesWeek of 2025-12-28: +2,522 linesWeek of 2025-12-28: -786 linesWeek of 2026-01-04: +7,675 linesWeek of 2026-01-04: -7,566 linesWeek of 2026-01-11: +22,141 linesWeek of 2026-01-11: -8,798 linesWeek of 2026-01-18: +156 linesWeek of 2026-01-18: -17 linesWeek of 2026-01-25: +1,591 linesWeek of 2026-01-25: -1,296 linesWeek of 2026-02-01: +30,146 linesWeek of 2026-02-01: -27,161 linesWeek of 2026-02-08: +104,603 linesWeek of 2026-02-08: -96,832 linesWeek of 2026-02-15: +245,774 linesWeek of 2026-02-15: -186,188 linesWeek of 2026-02-22: +20,541 linesWeek of 2026-02-22: -18,310 linesWeek of 2026-03-01: +2,801 linesWeek of 2026-03-01: -2,143 linesWeek of 2026-03-08: +13,850 linesWeek of 2026-03-08: -5,742 linesWeek of 2026-03-15: +109,280 linesWeek of 2026-03-15: -109,720 linesWeek of 2026-03-22: +66,372 linesWeek of 2026-03-22: -66,129 linesWeek of 2026-03-29: +51,047 linesWeek of 2026-03-29: -47,029 linesWeek of 2026-04-05: +3,041 linesWeek of 2026-04-05: -1,520 linesWeek of 2026-04-12: +9,949 linesWeek of 2026-04-12: -6,753 linesWeek of 2026-04-19: +129,960 linesWeek of 2026-04-19: -128,092 linesWeek of 2026-04-26: +65,023 linesWeek of 2026-04-26: -63,243 linesWeek of 2026-05-03: +74,054 linesWeek of 2026-05-03: -72,758 linesWeek of 2026-05-10: +310,962 linesWeek of 2026-05-10: -282,096 linesWeek of 2026-05-17: +113,092 linesWeek of 2026-05-17: -111,014 linesWeek of 2026-05-24: +43,009 linesWeek of 2026-05-24: -20,271 linesWeek of 2026-05-31: +56,341 linesWeek of 2026-05-31: -63,217 linesWeek of 2026-06-07: +55,324 linesWeek of 2026-06-07: -112,781 linesWeek of 2026-06-14: +9,832 linesWeek of 2026-06-14: -4,923 linesWeek of 2026-06-21: +16,620 linesWeek of 2026-06-21: -10,122 linesWeek of 2026-06-28: +2,674 linesWeek of 2026-06-28: -2,818 linesWeek of 2026-07-05: +62,742 linesWeek of 2026-07-05: -22,177 linesWeek of 2026-07-12: +49,780 linesWeek of 2026-07-12: -5,092 linesWeek of 2026-07-19: +290,492 linesWeek of 2026-07-19: -257,264 linesWeek of 2026-07-26: +60,785 linesWeek of 2026-07-26: -4,924 linesWeek of 2026-08-02: +78,553 linesWeek of 2026-08-02: -63,489 linesWeek of 2026-08-09: +152,423 linesWeek of 2026-08-09: -24,829 linesWeek of 2026-08-16: +43,366 linesWeek of 2026-08-16: -15,394 linesWeek of 2026-08-23: +51,933 linesWeek of 2026-08-23: -45,753 linesWeek of 2026-08-30: +38,516 linesWeek of 2026-08-30: -23,646 linesWeek of 2026-09-06: +84,053 linesWeek of 2026-09-06: -73,585 linesWeek of 2026-09-13: +36,956 linesWeek of 2026-09-13: -16,303 linesWeek of 2026-09-20: +87,404 linesWeek of 2026-09-20: -8,569 linesWeek of 2026-09-27: +50,525 linesWeek of 2026-09-27: -6,599 linesWeek of 2026-10-04: +49,032 linesWeek of 2026-10-04: -3,839 linesWeek of 2026-10-11: +20,280 linesWeek of 2026-10-11: -2,575 linesWeek of 2026-10-18: +24,907 linesWeek of 2026-10-18: -1,870 linesWeek of 2026-10-25: +43,645 linesWeek of 2026-10-25: -5,081 linesWeek of 2026-11-01: +47,707 linesWeek of 2026-11-01: -2,194 linesWeek of 2026-11-08: +119,111 linesWeek of 2026-11-08: -71,921 linesWeek of 2026-11-15: +170,919 linesWeek of 2026-11-15: -44,513 linesNov 23, 2025Nov 15, 2026
+3.2M lines added, -2.2M removed over the last year.

Commits per week

last 52 weeks
420Week of 2025-11-23: 8 commitsWeek of 2025-11-30: 5 commitsWeek of 2025-12-07: 0 commitsWeek of 2025-12-14: 3 commitsWeek of 2025-12-21: 2 commitsWeek of 2025-12-28: 6 commitsWeek of 2026-01-04: 2 commitsWeek of 2026-01-11: 3 commitsWeek of 2026-01-18: 2 commitsWeek of 2026-01-25: 4 commitsWeek of 2026-02-01: 14 commitsWeek of 2026-02-08: 10 commitsWeek of 2026-02-15: 16 commitsWeek of 2026-02-22: 8 commitsWeek of 2026-03-01: 13 commitsWeek of 2026-03-08: 13 commitsWeek of 2026-03-15: 14 commitsWeek of 2026-03-22: 10 commitsWeek of 2026-03-29: 4 commitsWeek of 2026-04-05: 6 commitsWeek of 2026-04-12: 11 commitsWeek of 2026-04-19: 20 commitsWeek of 2026-04-26: 10 commitsWeek of 2026-05-03: 11 commitsWeek of 2026-05-10: 12 commitsWeek of 2026-05-17: 17 commitsWeek of 2026-05-24: 29 commitsWeek of 2026-05-31: 24 commitsWeek of 2026-06-07: 25 commitsWeek of 2026-06-14: 18 commitsWeek of 2026-06-21: 12 commitsWeek of 2026-06-28: 8 commitsWeek of 2026-07-05: 21 commitsWeek of 2026-07-12: 23 commitsWeek of 2026-07-19: 21 commitsWeek of 2026-07-26: 16 commitsWeek of 2026-08-02: 26 commitsWeek of 2026-08-09: 27 commitsWeek of 2026-08-16: 39 commitsWeek of 2026-08-23: 24 commitsWeek of 2026-08-30: 40 commitsWeek of 2026-09-06: 35 commitsWeek of 2026-09-13: 23 commitsWeek of 2026-09-20: 39 commitsWeek of 2026-09-27: 42 commitsWeek of 2026-10-04: 34 commitsWeek of 2026-10-11: 20 commitsWeek of 2026-10-18: 19 commitsWeek of 2026-10-25: 35 commitsWeek of 2026-11-01: 28 commitsWeek of 2026-11-08: 40 commitsWeek of 2026-11-15: 25 commitsNov 23, 2025Nov 15, 2026
917 commits in the last 52 weeks.

When work happens

weekday and hour
SunMonTueWedThuFriSat036912151821Sun 0:00 — 11 commitsSun 1:00 — 8 commitsSun 2:00 — 3 commitsSun 3:00 — 6 commitsSun 4:00 — 3 commitsSun 5:00 — 3 commitsSun 6:00 — 2 commitsSun 7:00 — 3 commitsSun 8:00 — 1 commitsSun 9:00 — 6 commitsSun 10:00 — 4 commitsSun 11:00 — 5 commitsSun 12:00 — 10 commitsSun 13:00 — 12 commitsSun 14:00 — 13 commitsSun 15:00 — 9 commitsSun 16:00 — 8 commitsSun 17:00 — 11 commitsSun 18:00 — 16 commitsSun 19:00 — 10 commitsSun 20:00 — 8 commitsSun 21:00 — 10 commitsSun 22:00 — 16 commitsSun 23:00 — 21 commitsMon 0:00 — 6 commitsMon 1:00 — 4 commitsMon 2:00 — 2 commitsMon 3:00 — 5 commitsMon 4:00 — 4 commitsMon 5:00 — 1 commitsMon 6:00 — 2 commitsMon 7:00 — 3 commitsMon 8:00 — 3 commitsMon 9:00 — 2 commitsMon 10:00 — 4 commitsMon 11:00 — 2 commitsMon 12:00 — 3 commitsMon 13:00 — 5 commitsMon 14:00 — 9 commitsMon 15:00 — 8 commitsMon 16:00 — 5 commitsMon 17:00 — 9 commitsMon 18:00 — 6 commitsMon 19:00 — 8 commitsMon 20:00 — 14 commitsMon 21:00 — 11 commitsMon 22:00 — 17 commitsMon 23:00 — 27 commitsTue 0:00 — 16 commitsTue 1:00 — 12 commitsTue 2:00 — 4 commitsTue 3:00 — 4 commitsTue 4:00 — 5 commitsTue 5:00 — 5 commitsTue 6:00 — 4 commitsTue 7:00 — 1 commitsTue 8:00 — 1 commitsTue 9:00 — 4 commitsTue 10:00 — 7 commitsTue 11:00 — 1 commitsTue 12:00 — 7 commitsTue 13:00 — 4 commitsTue 14:00 — 6 commitsTue 15:00 — 6 commitsTue 16:00 — 7 commitsTue 17:00 — 9 commitsTue 18:00 — 9 commitsTue 19:00 — 9 commitsTue 20:00 — 14 commitsTue 21:00 — 13 commitsTue 22:00 — 17 commitsTue 23:00 — 9 commitsWed 0:00 — 9 commitsWed 1:00 — 6 commitsWed 2:00 — 3 commitsWed 3:00 — 2 commitsWed 4:00 — 2 commitsWed 5:00 — 7 commitsWed 6:00 — 1 commitsWed 7:00 — 3 commitsWed 8:00 — 4 commitsWed 9:00 — 1 commitsWed 10:00 — 1 commitsWed 11:00 — 0 commitsWed 12:00 — 8 commitsWed 13:00 — 3 commitsWed 14:00 — 3 commitsWed 15:00 — 7 commitsWed 16:00 — 7 commitsWed 17:00 — 9 commitsWed 18:00 — 7 commitsWed 19:00 — 11 commitsWed 20:00 — 15 commitsWed 21:00 — 16 commitsWed 22:00 — 14 commitsWed 23:00 — 15 commitsThu 0:00 — 11 commitsThu 1:00 — 12 commitsThu 2:00 — 1 commitsThu 3:00 — 5 commitsThu 4:00 — 1 commitsThu 5:00 — 3 commitsThu 6:00 — 6 commitsThu 7:00 — 5 commitsThu 8:00 — 6 commitsThu 9:00 — 0 commitsThu 10:00 — 1 commitsThu 11:00 — 4 commitsThu 12:00 — 4 commitsThu 13:00 — 2 commitsThu 14:00 — 3 commitsThu 15:00 — 6 commitsThu 16:00 — 6 commitsThu 17:00 — 7 commitsThu 18:00 — 4 commitsThu 19:00 — 5 commitsThu 20:00 — 14 commitsThu 21:00 — 14 commitsThu 22:00 — 13 commitsThu 23:00 — 17 commitsFri 0:00 — 8 commitsFri 1:00 — 8 commitsFri 2:00 — 5 commitsFri 3:00 — 7 commitsFri 4:00 — 2 commitsFri 5:00 — 2 commitsFri 6:00 — 1 commitsFri 7:00 — 2 commitsFri 8:00 — 2 commitsFri 9:00 — 1 commitsFri 10:00 — 3 commitsFri 11:00 — 1 commitsFri 12:00 — 0 commitsFri 13:00 — 2 commitsFri 14:00 — 2 commitsFri 15:00 — 3 commitsFri 16:00 — 5 commitsFri 17:00 — 11 commitsFri 18:00 — 10 commitsFri 19:00 — 8 commitsFri 20:00 — 10 commitsFri 21:00 — 10 commitsFri 22:00 — 12 commitsFri 23:00 — 19 commitsSat 0:00 — 11 commitsSat 1:00 — 8 commitsSat 2:00 — 4 commitsSat 3:00 — 2 commitsSat 4:00 — 0 commitsSat 5:00 — 3 commitsSat 6:00 — 3 commitsSat 7:00 — 5 commitsSat 8:00 — 1 commitsSat 9:00 — 5 commitsSat 10:00 — 8 commitsSat 11:00 — 4 commitsSat 12:00 — 7 commitsSat 13:00 — 5 commitsSat 14:00 — 8 commitsSat 15:00 — 9 commitsSat 16:00 — 7 commitsSat 17:00 — 19 commitsSat 18:00 — 11 commitsSat 19:00 — 10 commitsSat 20:00 — 10 commitsSat 21:00 — 11 commitsSat 22:00 — 15 commitsSat 23:00 — 16 commits
Commit volume by weekday and hour (UTC). Larger dots mean more commits.

Who is committing

last 52 weeks
Maintainer commits841 (89%)
Community commits101 (11%)

942 commits in total over the last year.

DateListRankStars gained
Sep 24, 2026daily#13+86
Sep 23, 2026daily#13+86
  • tensorflow/tensorflow

    An Open Source Machine Learning Framework for Everyone

    201.7K stars · C++

  • ggml-org/llama.cpp

    LLM inference in C/C++

    135.9K stars · C++

  • debpalash/VoiceStudio

    VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.

    115.5K stars · Python

  • Developer-Y/cs-video-courses

    List of Computer Science courses with video lectures.

    84.9K stars

  • rohitg00/ai-engineering-from-scratch

    Learn AI Engineering! Learn it. Build it. Ship it for others.

    79.6K stars · Python

  • ocornut/imgui

    Dear ImGui: Bloat-free Graphical User interface for C++ with minimal dependencies

    77.2K stars · C++