Each pixel becomes a square point, pushed back to its depth along the capture camera's ray. From the capture view the frame reads whole; any other view opens gaps. Depth Anything maps are near-is-bright, which is the default here.
The sample is one of my own photographs, Kabukichō in Tokyo, with depth from Video Depth Anything run locally. Drop in any frame and its depth map; with no depth map, depth falls back to blurred luminance, which only works as a toy.