Your engine bay,
from photographs.
A sparse 3D reference recovered from overlapping owner photographs. Rotate the points, or choose a source view and compare their projections with its photo. Engine and body observations now share a relative coordinate frame.
This reconstruction has arbitrary units. A measured length between identifiable reconstructed points is needed for physical scale; corresponding body datums are also needed for chassis alignment. These dots are observations, not closed surfaces or a 1:1 collision model.
Loading the photographic reference…
Drag to orbit; scroll or pinch to zoom. Showing a source photo limits the dots to features observed in that image. Orbiting returns to the full point set. Photographs are displayed unchanged.
What this adds
Fourteen existing photos now connect in one reconstructed bay fragment, including engine, front-body and rear-bay surroundings. The solve uses visible image features and estimated cameras, rather than assigning positions from an exploded drawing. Six other supplied images did not register into this main fragment; disconnected reconstructions are kept separate.
Original camera metadata distinguishes the main lens from two other-lens close-ups, and records two capture sessions. Shared visible features across those sessions do not establish unchanged hidden components or the car’s current state. The current CAD still needs an installed datum chain before this relative frame can justify moving coupled assemblies.
How to use it
Use the source views to inspect the observed surroundings and the CAD reference to explore component relationships. Missing points do not prove empty space. This data does not establish fitted clearances, tool access, working movement, replacement compatibility or a verified removal sequence.
Reconstruction evidence and limits
The main reconstruction registers 14 of 20 supplied images and contains 2,805 sparse points. The displayed subset has 2,305 points observed in at least three distinct images, with a maximum triangulation angle of at least five degrees and mean image error no greater than two analysis pixels. This filter removes weak tracks; it does not certify dimensional accuracy.
A separate per-image camera solve shares five front-bay images. Comparing corresponding tracks between the two pipelines gives a median difference of about 0.36% of their arbitrary reference span, and a 95th-percentile difference of about 1.66%. Both solves use the same photos and differ in feature extraction; this is sensitivity evidence, not independent survey validation. It does not establish millimetre accuracy.
Shared camera groups, a centered principal point, one radial distortion term and unchanged full-frame JPEG correspondence remain assumptions. Rounded original EXIF focal lengths are initial priors. The data uses PyCOLMAP’s sparse reconstruction; processing options, input fingerprints, camera poses and observation tracks are retained below.