Cycle Consistency as Reward: Learning Image-Text Alignment without Human Preferences

Open in new window