The following will create a new environment and install all dependencies required for SiLK.
make conda_env_init
conda activate silk
make dev_install
make conda_update
Then, assets (cached datasets, pre-trained model, ...) have to be downloaded using
./bin/public-pull-assets
Asset file which are not required for download can be removed from the assets/public-assets.txt
file before running the above command.
The assets are structured as such.
assets/datasets
contains datasets metadata, and cached hpatches datasets.assets/models
contains the model checkpoints used for all experiments mentioned in the paper.assets/models/silk/analysis/alpha/pvgg-4.ckpt
is the default model weights.assets/models/silk/analysis/*/
folders correspond to different result tables from the paper.assets/models/superpoint/*.ckpt
contains our reproduced SuperPoint model weights.
assets/results
contains json files of the results of every experiments mentioned in the paper.assets/tests
are files required to run the unit tests and check procedures.
Once the environement is ready, please run the following to test if things works fine.
./bin/run_tests
./bin/run_check_procedures