janusch
janusch
As the bonus challenge in bounty #284 was not cleared, I am offering a $500 payout for someone who can take both approaches, combine them, and clearly demonstrate an improvement...
Can this create useful poses? Is it possible to integrate the code? If so, how? https://github.com/93won/lightweight_vio
## Overview Implement interactive editing capabilities for Gaussian splats, bringing editing functionality similar to SuperSplat into our viewer. ## Proposed Editing Features ### 1. Selection Tools - **Box Selection**: Draw...
Directly digest it without converting first to Colmap format!/
See the title. Please investigate and fix if necessary.
There are a lot of cmake flags. It should be certainly possible to considerable reduce the binary size of libtorch so that we can create smaller binaries. Goal of this...
Currently the splats are not transformed. It is a fake view matrix transform that you can control with the world transform panel. This has to be implemented properly and fast...
This should go into the repo to improve adc densification: https://xiaobin2001.github.io/improved-gs-web/ Left mcmc + bilagrad, right improved-gs. Both with 2m Gaussians. https://github.com/user-attachments/assets/ffffdf80-f16e-4d41-bdaf-0b9ce501be4f
Pre-computes undistortion maps during camera setup (similar to OpenCV's approach but without the dependency) in colmap_reader.cpp. Updates camera intrinsics to account for the undistorted image geometry Reference: https://github.com/nerfstudio-project/gsplat/blob/main/examples/datasets/colmap.py