InstantSplat is a sparse-view, SfM-free framework for large-scale scene reconstruction method using Gaussian Splatting. It allows for extremely fast reconstruction by using Dust3r, a multiview stereo network, to initialize camera poses and dense point cloud for all training views. To enhance pose accuracy and tune scene parameters a gradient-based joint optimization framework from self-supervision is used. By employing this simplified framework, InstantSplat achieves a substantial reduction in training time, from hours to mere seconds, and demonstrates robust performance across various numbers of views in diverse datasets
This is an external example. Check the repository for more information.
You can try the example on HuggingFace space here.
It is highly recommended to run this example locally by cloning the above repo and running (make sure you have Pixi installed):
git clone https://github.com/pablovela5620/InstantSplat.git
pixi run app