SPAG4d Delivers PaGeR for Jitter-Free 360 Scenes
SPAG4d now offers the PaGeR generator, a depth-based backend that produces the most geometrically consistent 3D Gaussian splats from a single 360‑degree panorama. The tool turns any equirectangular photo into an explorable scene you can spin through in a web browser. By adding PaGeR, SPAG4d tackles the jittery alignment problems that often appear when stitching cube‑face depth maps together.
Developer cedarconnor pulled the PaGeR model, originally published by academic researchers, into the open‑source SPAG4d project as a one‑click option. The update means anyone with a suitable NVIDIA GPU can create cleaner outdoor 3D scenes directly from the app’s web UI. It focuses on solving multi‑face seam artifacts without sacrificing depth accuracy.
PaGeR brings stronger geometric consistency
What PaGeR adds
- Learned sky mask avoids outlier depth fits.
- World‑frame surface normals for cleaner edges.
- Auto indoor/outdoor detection using a CLIP router.
- Non‑commercial weights, needs about 12 GB VRAM.
Architectural visualizers, landscape photographers, and VR content creators will feel the biggest difference. A single 360‑degree photo now yields a walkable 3D scene with far fewer wobbly walls or floating artifacts, saving the multi‑camera hassle. The whole workflow runs in the familiar browser UI, so no command‑line experience is required.
What developers should keep in mind
PaGeR’s weights carry a CC BY‑NC 4.0 license, meaning you cannot use this backend for commercial deliverables. The generator needs roughly 12 GB of VRAM and an extra 5.7 GB model download before the first run, so it is heavier than the MIT‑licensed DA360 default. A pinned upstream commit keeps the integration stable, while the core pipeline remains MIT for commercial‑safe work with other backends.
“Geometrically consistent 360-degree scenes from single panoramas. Best alignment I’ve seen.” — Source: Reddit
Project Links
