Have Fun Storming the Castle(s)!

This is the official code repository for our WACV 2021 paper.

Dataset

Images	Castles	Countries	Construction Dates
772,927	2412	93	85% coverage

Download

We've provided a python script to download the data. The download process requires ~100 GB of disk space. Run the download script:

python download.py

This will download several file fragments and recompile them into a single castles.tgz file, checking md5 hashes along the way. Once the process is complete, you can remove the fragments:

rm *.tgz_fragment*

Then extract the data from castles.tgz:

tar xzf castles.tgz

Additional metadata, such as location and construction dates, can be found in metadata.json.

If you prefer to do the download manually, check download.py for remote urls and expected md5 hashes.

Note

The data is provided for research purposes. We do not own the rights to the images.

Experiments

All experiments use PyTorch version 1.6.0 and PyTorch Lightning version 0.9.0. Image retrieval experiments also depend on PyTorch Metric Learning version 0.9.93.