opendatalab/clip-parrot-bias

ECCV2024_Parrot Captions Teach CLIP to Spot Text

clip-parrot-bias's Issues

Why so many faces?

Hi, nice work!

I noticed that the example samples shown on HuggingFace mostly consist of human faces. Does the actual distribution of the dataset like this? If so, does this result in significant bias?

Invalid URLs within the datasets

Hi,

Thank you for releasing the dataset! When working on the dataset, I found that it contains lots of invalid URLs that do not link to valid images. For example,

ID: 3,034,453,019,914 , URL: https://www.belfasttelegraph.co.uk/incoming/article38085824.ece/9d030/AUTOCROP/w620h342/S_Johnny_Ronan_walk.jpg

ID: 3,329,324,007,388, URL: https://pics.me.me/thumb_adtak-oh-look-at-me-im-ryan-gosling-i-have-perfect-21541460.png

ID: 3,329,324,007,388, URL: https://pics.me.me/thumb_adtak-oh-look-at-me-im-ryan-gosling-i-have-perfect-21541460.png

ID: 3,763,989,082,165, URL: https://img.sharetv.com/shows/characters/thumbnails/a_million_little_things.gary.jpg

If you have all images stored locally, could yo release a version of the dataset with actual images instead of the URLs?

Thanks,
Jiachen