cvg/vidmap ? reverse-engineered prompt

Build me a Python tool that takes a video or a folder of extracted frames and turns it into a sparse 3D reconstruction.

I want it to run in two steps, first analyze the video to find tracks, keyframes, depth, and camera info, then run global mapping to estimate camera poses, intrinsics, and a final COLMAP style reconstruction. It should also support running everything in one command from input to output.

Please include a simple command line flow, sensible default settings, and a way to tweak the important reconstruction parameters if needed. I’d like the output folder to contain the final reconstruction plus saved config files. If possible, add optional visualization output too, like an interactive HTML view and a playback file for inspecting the reconstruction process.

Use current docs online if you need to, and make it work on Linux with Python.