each image has two views in our pre-training, i.e, image patches (such as 16x16 pixels), and visual tokens (i.e., discrete tokens).
makabaka2024/9/11
source
TermsPrivacy PolicyContact