Structure-from-Motion with LightGlue

This Github repository features the code for a Streamlit application for performing Structure-from-Motion (SfM) reconstructions using the LightGlue feature extractor. The application allows users to upload their own image datasets, select feature matchers, and generate 3D point cloud reconstructions.

For more information on LightGlue, please refer to their GitHub page.

Deploy on Google Colab for GPU

Structure-from-Motion is quite computationally intensive, and so, it is recommended that you leverage Google Colab's free GPU to boost the application. You can do this by:

1. Cloning the colab-sfm-w-lightglue python notebook into Colab and run all the cells.
2. Once all cells are running, you will see a link in the last cell that looks something like this: https://five-parents-crash.loca.lt.
3. Navigate to the local tunnel website and use the tunnel password in the colab notebook.