Unimodal embedding from text
#3
by
anselkthomas
- opened
Is there a way to create unimodal embedding from text with the same embedding space as the cross model embeddings?
Hi,
You can create unimodal embeddings from text by running the text through BridgeTowerForContrastiveLearning using any image and access the text embeddings with outputs.text_embeds.
See here for reference: BridgeTowerContastiveOutput