PromptEcho extracts annotation-free rewards from frozen VLMs via cross-entropy loss on the original prompt to improve prompt following in T2I models, with gains shown on DenseAlignBench and other benchmarks.
Pay attention to distinguishing Chinese/English, uppercase/lowercase, numbers, etc
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
PromptEcho: Annotation-Free Reward from Vision-Language Models for Text-to-Image Reinforcement Learning
PromptEcho extracts annotation-free rewards from frozen VLMs via cross-entropy loss on the original prompt to improve prompt following in T2I models, with gains shown on DenseAlignBench and other benchmarks.