The original Stonehenge video is here: https://www.youtube.com/watch?v=UrLUfDhHe-A
1) I use the yt-dlp package for downloading the youtube video (1440p)
2) used ffmpeg to extract frames at 1 fps, totaling 534 frames.
3) used colmap 3.13dev0+glomap to create poses for the images. It uses vocab_tree_matcher and detected 531 images. Finally I undistorted it with colmap image_undistorter
4) Did special training-preprocessing (dense cloud+remove blur)
5) I used gsplat simple_trainer for 30k and 2 M splats.
6) imported the gaussian splat to supersplat, cleaned it, and publish it with SOG