TYPECV / SFＭ LAB
Reconstruct a scene from video
Choose a local video to inspect frame sampling and calibration assumptions. The end-to-end GPU feature-to-SfM pipeline is still under development; this page reports that boundary clearly.
Video input
Uses an assumed focal-length/width ratio of 1.2, not calibrated camera metadata.
Frame features, BRIEF matching and geometric reconstruction run through TypeShade WebGPU. Three.js renders the result.
Choose a video and extract frames to begin.
Sample: Blanka OG, Drone Footage of a Pyramid on Pexels · Pexels License