Generation of Scene Graph and Semantic Image: A Review and Challenge Ahead

Scene graph generation creates a structured representation of visual scenes by identifying objects and their attributes and the relationships between them. Conversely, semantic image generation converts semantic representations such as scene graphs, textual descriptions, or object layouts, into photorealistic images. This paper provides an overview of recent advancements in generations of scene graph and semantic image. The survey explores how innovations in both fields can complement each other, paving the way for advanced applications in autonomous systems, semantic communication, visual reasoning, and creative industries. Challenges such as data imbalance, scalability, and semantic coherence and future research directions including multimodal integration, prior processing etc. are throughly discussed.