Datasets:

ArXiv:
License:

Is FaceCaption-1M a random subset of FaceCaption-15M?

#1
by melihsefa - opened

Hi,

First of all, thank you for releasing the FaceCaption dataset.

I have a question regarding the FaceCaption-1M release.

Could you clarify how the 1M dataset was constructed?

Specifically, I would like to know:

Is FaceCaption-1M simply a random sample of FaceCaption-15M?
Or was it created using additional filtering or quality selection?
If it is not a random subset, what criteria were used to select the 1M samples?

Sign up or log in to comment