Thirty seconds of ordinary phone footage, rebuilt in 3D with no LiDAR.
What this shows: roadside, kerbs, lane markings and the road surface come out in the right places and at the right scale. The camera sits 1.34 m above the fitted road, matching a windscreen mount, and the carriageway measures ~7.3 m, two 3.5 m lanes. What it doesn't show yet: the "staircase" edges are where neighbouring frames' depth maps disagree, from the ~10% error of one-camera depth. Stabilisation off, a calibrated lens, a higher frame rate, or a second phone for stereo would tighten them. Moving cars are left out of the fused map by design, since one camera can't triangulate things that move: each would leave a 66-copy ghost trail. "Vehicles at this frame" shows them as a LiDAR viewer would, one sweep at a time. Each box sits where the car's wheels meet the road, with its height measured (median 1.74 m) and its length and width from a class standard (box width can't separate a car's back from its side). Road defects come from Brevity's defects detector (patches and cracks at confidence 0.5 or more; potholes, rarer, at 0.3 or more), projected flat onto the fitted road. Positions are good to a few metres along the road: the same pothole seen from two frames lands about 5 m apart. Outlines longer than 15 m are left out: they come from boxes reaching towards the horizon, where a few pixels span tens of metres. Pothole depth can't be measured yet, since the ~10% depth error is far bigger than a pothole is deep. Stereo or closer frames would change that.