Spirula Studio v2026.10.30 can now orient and scale a reconstruction from AprilTag and ArUco markers placed in the scene.
Until this release, metric scale in Spirula Studio's gaussian splatting reconstructions came from GPS or from laser scans, which v2026.10.6 brought into training. Version 2026.10.16 followed with RoMa v2 dense initialization two weeks ago.
The marker tool, labeled experimental, detects AprilTag tag36h10 and tag36h11 or ArUco 4x4 to 7x7 markers in the dataset images and triangulates each one from the camera poses. Three markers define the origin, the positive X axis and the ground plane, and the measured distance between the first two sets metric scale. The transform is baked into the COLMAP reconstruction and the trained splat PLY, and a `spirula marker` command can also draw a crop box around the markers. Without markers, the editor can scale a model from the known distance between two picked points.
Depth and normal estimation adds MoGe 3, which runs a sparse 3D refiner after the MoGe 2 network for sharper depth edges, and Marigold v2. Dataset creation can skip blurry and near duplicate frames, with a Sharpness view to review the choice, and a match matrix editor removes false matches by hand. Spirula Studio now also parses masks named in the RealityScan convention, reads a GPS origin from gauge.txt, and splits QooCam 3 Ultra raw footage into two fisheye lenses with IMU data.
Training changes target web viewers and quality after compression, with better compatibility with PlayCanvas and Spark. They ask for issues on any regression, without published metrics. Evaluation now reports LPIPS, and the compare view gains a split slider wipe across two to four models.
Structure from motion gains RAM and VRAM forecasting. Fixes cover older AMD GPUs and occasional out of memory errors from VRAM spikes during training.
Spirula Studio v2026.10.30 is available now on GitHub.
