Jonathan Astermark · Anders Heyden · Viktor Larsson

1. Create virtual environment and install dependencies

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

2. Install the dms package, as well as pysumopt (which contains the pybinds)

git submodule update --init --recursive
cd cpp && pip install --no-build-isolation --no-deps . && cd ..
pip install --no-build-isolation --no-deps -e .

The provided notebook demo.ipynb contains a small demo for estimating the relative pose from a single image pair. To run the demo, you need the MegaDepth-1500 images and DKM-matches. You can download both using the provided shell script data/download_megadepth.sh:

cd data
bash download_megadepth.sh
cd ..

We also provide a notebook run_benchmarks.ipynb for running full evaluation on either MegaDepth-1500 or ScanNet-1500, using any matches (DKM and RoMa-matches are provided for download) Note: Before running the benchmarks, you need to run clustering using the provided script run_clustering.py. Example usage:

python run_clustering.py --dataset megadepth --keypoints roma --method kmeans4d --num_components 128

@inproceedings{astermark2025dense,
  author    = {Astermark, Jonathan and
               Heyden, Anders and
               Larsson, Viktor},
  title     = {Dense Match Summarization for Faster Two-view Estimation},
  booktitle = {Computer Vision and Pattern Recognition (CVPR)},
  year      = {2025}
}