I am trying out some stuff with the mask file and the COLMAP data structure.
We know that creating spatially connected masks in 3D space is difficult because the masks are disconnected from each other, there are occlusions etc. I have however managed to group masks together masks based on the spatial sparse data point cloud, even with occlusions and complex environments. It’s not the best, but works actually pretty good. Might miss a few masks, but generally groups them pretty good. If no matches are found, those masks are discarded and grouped separately.
The code doesn’t however create any Gaussians, this is just pure grouping algorithmic approach to COLMAP and mask files.
Is that something that would be of interest for you and can try out with Gaussian yourself?
Hi, this is interesting and may help refresh the research in this area. Would you like to provide more details / visualization results about your work? We may discuss about it deeper. You may reach me by e-mail.
Hello,
I am trying out some stuff with the mask file and the COLMAP data structure.
We know that creating spatially connected masks in 3D space is difficult because the masks are disconnected from each other, there are occlusions etc. I have however managed to group masks together masks based on the spatial sparse data point cloud, even with occlusions and complex environments. It’s not the best, but works actually pretty good. Might miss a few masks, but generally groups them pretty good. If no matches are found, those masks are discarded and grouped separately.
The code doesn’t however create any Gaussians, this is just pure grouping algorithmic approach to COLMAP and mask files.
Is that something that would be of interest for you and can try out with Gaussian yourself?