This project just modifies the official gaussian splatting code to jointly optimize camera poses alongside the splat
In practice, this is just a slight refinement to the poses for more pixel-perfect alignment rather than a robust pose optimization
The code is not very polished but we hope it's useful
Just point your colmap-formatted scene as follows: python train.py -s /nobackup/flowmap_results/colmap/tandt_caterpillar --name test_caterpillar -o and it should launch a wandb run for the optimization
The environment to run gaussian splatting is a bit tricky but you should start with getting the official splatting code running (https://github.com/graphdeco-inria/gaussian-splatting) and then come here after and it should just work