SSS360

From 360° video to clean photogrammetry plates for 3D Gaussian Splatting — automatic AI masking of everything you don't want in the scan.

Python PyTorch SAM CUDA RealityCapture OpenCV

What it does

SSS360 takes the equirectangular video from a 360° camera and prepares it for photogrammetry. It runs AI segmentation (SAM) on the footage to automatically detect and mask out everything that shouldn't end up in your 3D reconstruction — sky, power lines, rigs, vehicles and people — then reprojects the sphere into a set of overlapping flat views and packages them so they're ready to drop straight into RealityCapture.

How it works

Import & pick a frame. Load the 360° video and preview any frame instantly to check framing before processing.
AI masking. SAM segments each equirectangular frame, masking sky, cables, the operator and other unwanted objects — with manual control over dilation, smoothing and the nadir (bottom) cover.
Reprojection. The spherical frame is reprojected into overlapping flat views (a configurable ring + top/bottom) so RealityCapture sees clean, undistorted images.
Auto-packaging. Views are batched and sized to your available RAM, with quality auto-adapted to the source resolution (up to 8K), ready for RealityCapture → 3D Gaussian Splats.

Built for

Street & environment capture from a single 360 camera
Removing sky, cables & the operator automatically
Feeding RealityCapture → Gaussian Splats / 3D scenes
High-resolution footage up to 8K equirectangular

Not for

Real-time / live video masking
Single flat photos (it expects 360° equirectangular)
PCs without an NVIDIA RTX / GTX 16xx GPU

Highlights

One-click pipeline with a clean desktop GUI — no command line.
Automatic sky & cable removal powered by SAM segmentation.
Smart RAM-aware packaging with adjustable overlap for solid alignment.
Auto-quality based on resolution + a RealityCapture preset for street capture.
Fast preview viewer with fluid zoom / pan to inspect masks at full detail.

Demo

Coming soon