Skip to content

Latest commit

 

History

History
47 lines (31 loc) · 1.13 KB

README.md

File metadata and controls

47 lines (31 loc) · 1.13 KB
Raw

Correspondence Retrieval Experiments

System Requirements

Installation

  1. For comparing sgd-based clustering against default clustering, we need a fast clustering package (faiss).
conda install -c pytorch faiss-gpu

Command

  1. Download the correspondence pair datasets.
python ./download_datasets.py

We provide google drive link to the extracted features of the datasets. The above script downloads all available datasets. Alternatively, you can build the necessary dataset yourself by running step 2.

  1. Run a dummy experiment.
cd code
python cli.py run

The above script will download original datasets for integrity check. Also, it will extract necessary features if they are not downloaded already in step 1.

  1. Run all experiments.
cd code
bash ./run.sh

To enable GPU computation, modify the CUDA_VISIBLE_DEVICES environment variable accordingly. For example, run the above command as export CUDA_VISIBLE_DEVICES=2,3; bash ./run.sh.

  1. Check the output at the ../data/correspondence_retrieval/output directory.