AINeutralarXiv – CS AI · 7h ago5/10
🧠
Envisioning Beyond the Few: Disentangled Semantics and Primitives for Few-Shot Atypical Layout-to-Image Generation
Researchers propose a novel framework for layout-to-image generation that improves visual quality in few-shot learning scenarios by disentangling semantic identity from visual details. The method uses semantic anchoring and primitive imbuing to address representation fragmentation, enabling more coherent image synthesis from sparse training data.