AesBench

AesBench is an expert benchmark designed to comprehensively evaluate the aesthetic perception capacities of Multimodal Large Language Models (MLLMs) when it comes to image aesthetics perception. Let me break it down for you:

1. Purpose and Challenge:
2. MLLMs, which combine language and vision, are rapidly advancing.
3. However, their performance in aesthetic perception (assessing the beauty or visual appeal of images) remains uncertain.

4. The lack of a specific benchmark for evaluating MLLMs in this domain hinders their further development.

5. What Is AesBench?:

6. AesBench addresses this challenge by providing a comprehensive benchmark.

7. It evaluates MLLMs' aesthetic perception abilities through dual facets:

   • Expert-labeled Aesthetics Perception Database (EAPD): This database contains diverse image contents with high-quality annotations from professional aesthetic experts.
   • Integrative Criteria: AesBench proposes criteria to measure MLLMs' aesthetic perception abilities from four perspectives:
   • Perception (AesP): How well MLLMs perceive aesthetics.
   • Empathy (AesE): Their ability to empathize with aesthetic preferences.
   • Assessment (AesA): How accurately they assess aesthetics.
   • Interpretation (AesI): Their understanding of aesthetic features.

8. Findings:

9. Extensive experiments reveal that current MLLMs possess only rudimentary aesthetic perception ability.
10. There remains a significant gap between MLLMs and human aesthetic perception.

In summary, AesBench provides a valuable tool for assessing how well MLLMs understand and appreciate the beauty of images. 📸🌟

(1) [2401.08276] AesBench: An Expert Benchmark for Multimodal Large .... https://arxiv.org/abs/2401.08276. (2) GitHub - yipoh/AesBench: An expert benchmark aiming to comprehensively .... https://github.com/yipoh/AesBench. (3) AesBench/README.md at main · yipoh/AesBench · GitHub. https://github.com/yipoh/AesBench/blob/main/README.md. (4) undefined. https://doi.org/10.48550/arXiv.2401.08276.