TGO Aligns Visual Generative Models with Scalar Feedback Without Preference Pairs | ICML 2026
无需构造偏好对:TGO用标量反馈对齐视觉生成模型|ICML'26
NUS proposed Threshold-Guided Optimization, which converts scalar feedback into positive or negative updates through a score-distribution threshold and was accepted by ICML 2026; experiments cover Stable Diffusion v1.5, FLUX, Wan 1.3B, and Meissonic across image and video generation settings.
Why it matters: NUS's TGO, accepted at ICML 2026, skips preference-pair construction and aligns visual generative models directly from scalar scores like aesthetics or image-text matching by thresholding.