CoCa: Contrastive Captioners Are Image-Text Foundation Models (2022)
This is a 2022 paper from Google. For me, the key idea is captured in the following passage and two diagrams.
Training combines annotated images with noisy image–text pairs. The class labels in
