Evaluating Text-to-Image Synthesis:
Survey and Taxonomy of Image Quality Metrics

Sebastian Hartwig1, Dominik Engel1, Leon Sick1, Hannah Kniesel1, Tristan Payer1, Poonam1, Michael Glöckler1, Alex Bäuerle2, Timo Ropinski1
1Ulm University 2Carnegie Mellon University
Paper arXiv
Interpolate start reference image.

Evaluating text-to-image synthesis models includes two types of quality measures that contribute to an overall image quality. Compositional Quality measures how well the image reflects the composition defined in the text prompt. General Image Quality measures the overall quality of the image. For both types, several different aspects can be considered, e.g. realism might be important only for some applications. After measuring each aspect of the two categories, an aggregated quality score can be computed.

Abstract

Recent advances in text-to-image synthesis enabled through a combination of language and vision foundation models have led to a proliferation of the tools available and an increased attention to the field. When conducting text-to-image synthesis, a central goal is to ensure that the content between text and image is aligned. As such, there exist numerous evaluation metrics that aim to mimic human judgement. However, it is often unclear which metric to use for evaluating text-to-image synthesis systems as their evaluation is highly nuanced. In this work, we provide a comprehensive overview of existing text-to-image evaluation metrics. Based on our findings, we propose a new taxonomy for categorizing these metrics. Our taxonomy is grounded in the assumption that there are two main quality criteria, namely compositionality and generality, which ideally map to human preferences. Ultimately, we derive guidelines for practitioners conducting text-to-image evaluation, discuss open challenges of evaluation mechanisms, and surface limitations of current metrics.

BibTeX

@article{hartwig2024evaluating,
      title={Evaluating Text-to-Image Synthesis: Survey and Taxonomy of Image Quality Metrics}, 
      author={Sebastian Hartwig and Dominik Engel and Leon Sick and Hannah Kniesel and Tristan Payer and Poonam Poonam and Michael Glöckler and Alex Bäuerle and Timo Ropinski},
      year={2024},
      eprint={2403.11821},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}