数据集:
laion/laion400m
许可:
This datasets has two improvements compared to original LAION_400m dataset:
All in all, we filtered out around 6 million additional image-text pairs - probably with a high false positive rate - in order to improve dataset safety.