News

The text input is processed by the text encoder, while the image undergoes our patch generation-to-selection strategy before entering the image encoder. The loss subsequently aligns the visual and ...