I have a high-resolution video of a scene and want to turn it into a 3D splat reconstruction that opens cleanly in any standard Gaussian-splat viewer. The pipeline must be written in Python and should: • Ingest the video (or a set of images, two option should be proposed : input a video or a path of a 3D folder with images) of a scene directly (no camera intrinsics/extrinsics will be supplied). • Estimate all necessary camera parameters on the fly, reconstruct the geometry, and produce a coloured point-cloud splat. • Export two files: scene.ply and scene.splat. A command-line interface with clear arguments is enough; please document any external tools or libraries (GPU if needed) the script relies on and include a short README so I can reproduce the result on my own machine.