Kakao Brain releases open-source ViT and ALIGN models with COYO dataset

AI Impact Summary

Kakao Brain has released open-source versions of the ViT and ALIGN models, along with the COYO dataset, offering a significant advancement in visual language modeling accessibility. These models, trained on the 700 million image-text pairs in the COYO dataset, achieve performance comparable to Google's proprietary ViT and ALIGN models, particularly when leveraging the open-source training data. This provides researchers and developers with a readily available, reproducible alternative for visual language tasks, reducing reliance on large, inaccessible datasets.