pdigger-outdoorsEXPERIMENTAL

BEAN / Spatial time explorer

Reconstructed day — observed events 23h 56m
3D WORLD SOURCE VIEW
SOURCE ELAPSED00:00:00.000Loading evidence
Source-derived geometry · relative scale
Building your spatial viewLoading reconstruction and observed tracks…
PerspGROUND AXES
00:00:00.000/ 23:56:00

COMPOSITION

— layers
LAYER VISIBILITY
ABOUT THIS RECONSTRUCTION

A day, reconstructed in space.

pdigger-outdoors is a real WebGL 3D viewer of the recovered camera world, observed object tracks and reconstructed mesh hypotheses. The optional reference frame is a separate comparison aid.

What is measured

Source-relative timestamps and image-space observations inside the recovered event windows. Filters do not imply day-long coverage.

What is estimated

Single-view depth, camera intrinsics, ground projection and 3D placement use relative scale. Hidden surfaces and metric dimensions are not established.

What is not inferred

Smooth display positions and headings use observed runs and explicitly reviewed same-object short gaps. Interpolated poses are labelled estimates, not added observations. Long or ambiguous gaps remain unplaced; stubs do not authorize inventing locations. Discrete points use a ±0.035s display tolerance, not an inferred duration. Repeated appearances are event IDs, not verified unique identities.

How to navigate

Drag to orbit, scroll to dolly, right-drag to pan. W/A/S/D translate; Q/E move vertically; Shift moves faster. R restores the source camera. World and Orthographic use level-ground viewing axes; they do not remove distortion embedded in the estimated geometry. Space plays or pauses. The upper-right XYZ gizmo snaps around the current orbit pivot; its center toggles projection. Y is the estimated ground normal. The ground grid is real 3D and follows the active camera projection, with a tiny rendering lift to avoid surface flicker. Source lens distortion has no calibrated coefficients, so no artificial fisheye warp is applied.

Download the evidence

Real source data, not a calibrated survey. Raw predictions load only if downloaded.

World geometry · GLBEstimated camera · JSONRefined track evidence · JSONOriginal track snapshotOriginal inferred world · GLBRaw depth & normals · ZIPInference receiptFrame provenanceMarigold licenseImage-fitted SAM3D calibrationSAM3D source bindingsExact body-pose provenanceAdditional source contextReviewed tire-contact annotationsType-stub / study model provenanceModel / animation notesLoad evidence limitsReviewed continuity bridgesReviewed waiting episodeImage-fitted type-stub estimatesWheel-preserving display correction

Passing-truck load review

Native source frames · coarse apparent fill ranges

Percentages estimate visible fill, not calibrated capacity, volume or weight. Loaded cutaways show a representative band fill; near-empty models expose the floor and any observed residual patch. Covered or occluded interiors stay unknown—not empty trucks. These source observations do not validate existing 3D reconstruction or locate an unseen truck.

Download review and provenance