Fine-grained Image Captioning with CLIP Reward.
Jaemin ChoSeunghyun YoonAjinkya KaleFranck DernoncourtTrung BuiMohit BansalPublished in: NAACL-HLT (Findings) (2022)
Keyphrases
- fine grained
- coarse grained
- image features
- image classification
- single image
- image content
- image data
- multiscale
- input image
- access control
- massively parallel
- image representation
- image retrieval
- image segmentation
- energy function
- low level
- databases
- tightly coupled
- probabilistic model
- reinforcement learning
- high level
- metadata