Every overlay here is real model output, computed offline on an official Stevens photo; the faculty photos are untouched, and the links open each page with motion on.
1The sheet sweeps in, far objects open first
2Objects open from far to near
3Every object outlined and labelled
4At rest, the relations drawn in red
Phone, first screen
The Babbio Center atrium sits under a light white sheet, and each object the models found is cut out of it in full colour, outlined and labelled with the detector’s score. Thin red lines join related objects and name what was measured from their outlines and depth, such as “lawn behind handrail”, and pointing at an object shows all of its relations.
Open the page for sample PA, Object Relations, with motion on
1The farthest segments first, 31 of 336 placed
2Upper glass and cladding, 127 placed
3Curtain wall and doors, 222 placed
4Canopy, benches and shrub, 315 placed
Phone, first screen
The Babbio Center entrance, the same photo as P, starts as a dark ground carrying the faint plan of the 336 segments SAM drew on it. The segments land from far to near in the order Depth Anything V2 gives them, most first as flat colour and then as the photo, and pointing at one reads out its place in the build, its depth and its size.
Open the page for sample PB, Built by the Model, with motion on
1The cloud, turned to show its depth
2Scroll: flying to the photo’s viewpoint
3Each dot back on its own pixel
4The photograph itself
Phone, first screen
The Babbio Center atrium as 110,000 dots (30,000 on a phone), each one pixel of the photo set at the distance Depth Anything V2 estimated for it and drawn in that pixel’s colour, seen a little from the side with lines to where the photo was taken. On load a depth wave pushes each dot forward or back out of a flat picture, the view turns slightly with the pointer, and scrolling flies back to the photo’s own viewpoint until the dots become the photograph.