CGPO adaptively increases sampling of prompts whose image groups show high reward variance, combined with category calibration, to improve text-to-image generation on GenEval, T2I-CompBench++, and DPG Bench.
2d-curri-dpo: Two- dimensional curriculum learning for direct preference opti- mization, 2025
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
-
Curriculum Group Policy Optimization: Adaptive Sampling for Unleashing the Potential of Text-to-Image Generation
CGPO adaptively increases sampling of prompts whose image groups show high reward variance, combined with category calibration, to improve text-to-image generation on GenEval, T2I-CompBench++, and DPG Bench.