A Comprehensive Survey of Scene Graphs: Generation and Application

Area of Science:

• Computer Vision
• Artificial Intelligence
• Data Representation

Background:

• Current computer vision focuses on object recognition, but higher-level scene understanding is needed.
• Scene graphs provide a structured representation of objects, attributes, and relationships within an image.
• Advanced tasks like visual relationship detection, image captioning, and visual question answering require deeper scene comprehension.

Purpose of the Study:

• To provide a systematic survey of current scene graph research.
• To define scene graphs and discuss their generation methods.
• To explore scene graph applications and common datasets.

Main Methods:

• Summarizing the definition of scene graphs.
• Discussing scene graph generation (SGG) methods, including those using prior knowledge.
• Investigating major applications and datasets for scene graphs.

Main Results:

• Scene graphs are powerful tools for scene understanding, bridging object recognition and higher-level reasoning.
• Research in scene graphs is cross-modal, complex, and rapidly evolving.
• A comprehensive review of SGG methods and applications is presented.

Conclusions:

• Scene graphs are crucial for advancing visual scene understanding beyond object detection.
• This survey offers a structured overview and insights into future directions for scene graph research.
• The field of scene graphs is dynamic, with significant potential for future development.