EUVIMEDEUVIMED entdecken
Crossref

Diff-IR: A Non-Autoregressive End-to-End Framework for Generative Image Retrieval via Diffusion Models

Jian-Xin Ren, Ming-Bo Zhao, Zhao Zhang, Jicong Fan, Han Yan, Xuejiao Zhao

IEEE Transactions on Consumer Electronics · 2026-05 · 72 (2) · 3789-3796

DOI: 10.1109/tce.2026.3686635 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Consumer Electronics, 2026-05. Verfasst von Jian-Xin Ren und 5 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: National Natural Science foundation, Key Scientific Research Platform and Project of Guangdong Provincial Education Department.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Diffusion models have recently shown remarkable generative and modeling capabilities across vision and audio domains. Motivated by this progress, we propose Diff-IR, a novel generative image retrieval framework that reformulates retrieval as a non-autoregressive sequence generation task. Instead of the conventional two-stage pipeline—feature extraction followed by similarity search—Diff-IR directly encodes each query image into a compact discrete sequence whose code uniquely matches database images. This end-to-end design eliminates explicit similarity computation. In Diff-IR, high-level semantics are first extracted via a pre-trained CLIP encoder. The query features and a noise sequence are then processed by a Transformer-based denoising network, which gradually reconstructs the target discrete representation. Experiments on three public benchmarks demonstrate that treating retrieval as a generative task with diffusion not only improves accuracy but also offers efficient, fully end-to-end retrieval. Our results highlight the strong potential of diffusion models for next-generation retrieval systems.

Förderung

  • National Natural Science foundation
  • Key Scientific Research Platform and Project of Guangdong Provincial Education Department

Zitieren

APA

Ren, J., Zhao, M., Zhang, Z., Fan, J., Yan, H., & Zhao, X. (2026). Diff-IR: A Non-Autoregressive End-to-End Framework for Generative Image Retrieval via Diffusion Models. IEEE Transactions on Consumer Electronics, 72(2), 3789-3796. https://doi.org/10.1109/tce.2026.3686635

Vancouver

Ren J, Zhao M, Zhang Z, Fan J, Yan H, Zhao X. Diff-IR: A Non-Autoregressive End-to-End Framework for Generative Image Retrieval via Diffusion Models. IEEE Transactions on Consumer Electronics. 2026;72(2):3789-3796. doi:10.1109/tce.2026.3686635

BibTeX
@article{Ren2026,
  author = {Ren, Jian-Xin and Zhao, Ming-Bo and Zhang, Zhao and Fan, Jicong and Yan, Han and Zhao, Xuejiao},
  title = {Diff-IR: A Non-Autoregressive End-to-End Framework for Generative Image Retrieval via Diffusion Models},
  journal = {IEEE Transactions on Consumer Electronics},
  year = {2026},
  volume = {72},
  number = {2},
  pages = {3789--3796},
  doi = {10.1109/tce.2026.3686635}
}
Kostenlos starten →