Evaluation Metrics for Intelligent Generation of Graphical Game Assets: A Systematic Survey-Based Framework

Area of Science:

• Computer Graphics
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Generative systems offer automated creation of graphical assets.
• Current evaluation methods lack standardization, hindering comparison and validation.
• A unified approach is needed for quantitative assessment of generative methods.

Purpose of the Study:

• To propose a framework for selecting appropriate metrics for evaluating generative graphical asset systems.
• To provide guidance for practitioners in validating and comparing different generative approaches.
• To establish benchmarks for future research in generative asset creation.

Main Methods:

• A comprehensive literature review of approximately 200 papers was conducted.
• A framework was developed based on the literature to guide metric selection.
• The framework considers asset validity, quality, and operational capabilities.

Main Results:

• The framework facilitates informed decisions on metric selection for diverse generative methods.
• It addresses the need for standardized evaluation in the field of graphical asset generation.
• Guidance is provided for assessing both the output artefacts and the system's performance.

Conclusions:

• A unified framework is crucial for the quantitative evaluation of generative graphical asset systems.
• The proposed framework aids in selecting relevant metrics for assessing asset quality and system performance.
• This work supports the advancement of generative AI for graphical asset creation through standardized evaluation.