Confidence-aware Reward Optimization for Fine-tuning Text-to-Image Models.
Kyuyoung KimJongheon JeongMinyong AnMohammad GhavamzadehKrishnamurthy DvijothamJinwoo ShinKimin LeePublished in: CoRR (2024)
Keyphrases
- fine tuning
- image data
- input image
- random fields
- image classification
- image regions
- single image
- multiscale
- bayesian framework
- template matching
- image features
- viable alternative
- high resolution
- image content
- region of interest
- segmentation method
- image segmentation
- test images
- image retrieval
- parametric models
- web images
- image representation
- statistical model
- fine tune
- information retrieval
- image analysis
- text information
- image statistics
- visual effects
- segmentation algorithm
- confidence scores
- similarity measure
- textual information
- reinforcement learning
- pixel values
- low level
- image registration
- image matching
- edge detection