EUVIMEDEUVIMED entdecken
Crossref

Image Quality Assessment for Machine-oriented Communication

Chunyi Li, Xiaoyue Ling, Farong Wen, Haoyang Li, Guo Lu, Weisi Lin, Xiao-Ping Zhang, Guangtao Zhai, Wenjun Zhang

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026 · 1-18

DOI: 10.1109/tpami.2026.3748556 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. Verfasst von Chunyi Li und 8 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid development of Multimoal Large Language Models (MLLMs) has driven the evolution of communication paradigms, making the transmission of visual information in extremely low-bandwidth possible. However, as it is oriented toward computer vision tasks rather than humans, there is no widely accepted metric for its evaluation. To address this issue, we conduct a comprehensive visual quality assessment of machine oriented communication systems. Specifically, we (1) at the source level: a generative codec framework is established, with a MLLM communication pipeline through cross-modal generation between images and text. (2) at the channel level: images are transmitted using channel encoding/decoding strategies in different physical environments, and construct a dataset with 12k pairs of reference/distorted images. (3) at the sink level: 50 computer vision models are implemented to conduct 600k inferences on 5 downstream tasks, to characterize the differences between refer ence/distorted image pairs. (4) at the evaluation level: benchmark 18 cutting-edge Image Quality Assessment (IQA) methods, and develop a novel Mixture-of-Experts metric, termed CFA-MoE, to serve as a strong baseline for this task. Experimental results show that the current IQA algorithms cannot accurately evaluate machine-oriented distortions, highlighting the importance and challenge of this task. It is expected this dataset and baseline to facilitate future IQA algorithms and promote the communication paradigms from human to machine preference. Code will be available at: https://github.com/aiben-ch/MoQAD.

Zitieren

APA

Li, C., Ling, X., Wen, F., Li, H., Lu, G., Lin, W., Zhang, X. P., Zhai, G., & Zhang, W. (2026). Image Quality Assessment for Machine-oriented Communication. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-18. https://doi.org/10.1109/tpami.2026.3748556

Vancouver

Li C, Ling X, Wen F, Li H, Lu G, Lin W, et al. Image Quality Assessment for Machine-oriented Communication. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026:1-18. doi:10.1109/tpami.2026.3748556

BibTeX
@article{Li2026,
  author = {Li, Chunyi and Ling, Xiaoyue and Wen, Farong and Li, Haoyang and Lu, Guo and Lin, Weisi and Zhang, Xiao-Ping and Zhai, Guangtao and Zhang, Wenjun},
  title = {Image Quality Assessment for Machine-oriented Communication},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2026},
  pages = {1--18},
  doi = {10.1109/tpami.2026.3748556}
}
Kostenlos starten →