Chameleon is a model based on tokens which can interpret and produce text and images in any order. Its entirely token-based structure allows for smooth data integration across different modalities. By turning images into distinct tokens and training from the ground up on mixed-modal data, Chameleon is able to collaboratively process both image and text in a completely novel manner.
Chameleon
Creator
Creator
Seonglae ChoCreated
Created
2024 May 22 4:56Editor
Editor
Seonglae ChoEdited
Edited
2024 Jul 17 15:23Refs
Refs
Meta FAIR