◉ SPHERE LAB LOCAL INFERENCE

From a photograph to a spatial view.

03 / SPATIAL VIEW

Point cloud

In VR: pinch and hold with one hand to grab and move. Pinch with both hands to scale and turn; each pinch stays anchored to its grab point. Release to leave the cloud in place. A quick double pinch opens the context menu; move up to highlight debug or config and release.

04 / RESULTS

Waiting for a photo

Scores are detector confidence values, not calibrated probabilities. Overlapping candidates are suppressed at IoU 0.75. Gold wireframes show fitted depth profiles before skew correction, mapped into display space; they are not metric measurements. Center depth uses reciprocal input units; depth scale converts those units to pixels for the fit.

CandidateScoreCenter (px)Radius (px)Center depth (relative)Depth scale (px / depth unit)
Results appear after the detector finishes.

Relative depth only · Assumed 60° horizontal field of view · Display scale is adjustable and does not represent measured distances.