we could already check for duplicate file-names in the crawling if you (optionally) pass a metadata csv, and then skip the files for which the file names are already in

we do have a metadata file which contains all the images that have already been downloaded. If we could pass this file to the crawler as an argument, we could find out which images are already in our collection. Having this information we could then put them in a separate "duplicate images" folder, so we do not have them twice in our annotations

the collection-specific metadata csvs still have the Pomander in it, where it should be the keyword searched for, e.g. "chimney". Can you try finding the place in the code where this is done and adapt it to the current keyword?

1. Think of possible (sub-)classes
2. Add classes to olfactory-phenomena project in CVAT
3. Add a short description of (sub-)classes in annotation guidelines

human

Manual upload seems a bit complicated and takes too much time

Please document your investigations here (even if it leads to zero results):
https://github.com/Odeuropa/wp2-annotations/blob/master/image-collection/gestures.md