EUVIMEDEUVIMED entdecken
Crossref

FaceCaption-15M: Benchmarking and Enhancing Facial Vision--Language Models

Dawei Dai, Yutang Li, Yingge Liu, Mingming Jia, Yuanhui Zhang, Shuyin Xia, Guoyin Wang

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 16711-16724

DOI: 10.1109/tcsvt.2026.3713342 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Dawei Dai und 6 weiteren Autor:innen. Die Arbeit nennt 76 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Chongqing.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Currently, image-text-driven deep learning models have demonstrated their outstanding potential in many fields. In practice, facial tasks have broad application prospects. This paper presents FaceCaption-15M, a large-scale, diverse, and high-quality dataset of facial images accompanied by their natural language descriptions. Our FaceCaption-15M comprises over 15 million pairs of facial images and their corresponding natural language descriptions of facial features, making it the largest facial image-caption dataset to date. We conducted a comprehensive analysis of image quality, text naturalness, text complexity, and text-image relevance to demonstrate the superiority of FaceCaption-15M. To validate the effectiveness of FaceCaption-15M, we first trained a facial language-image pre-training model (FaceLIP) to align facial image with its corresponding captions. Subsequently, using both image and text encoders and fine-tuning only the linear layer, our FaceLIP-based models achieved state-of-the-art results on three challenging face-centered tasks. We also developed a domain-specific text-to-image models for facial image generation, and the results demonstrate that our model can generate the desired and high-quality facial images. The purpose is to promote research of face-related field through the availability of the proposed FaceCaption-15M. The Data, codes, and models are publicly available at: https://huggingface.co/datasets/OpenFace-CQUPT/FaceCaption-15M

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Chongqing

Zitieren

APA

Dai, D., Li, Y., Liu, Y., Jia, M., Zhang, Y., Xia, S., & Wang, G. (2026). FaceCaption-15M: Benchmarking and Enhancing Facial Vision--Language Models. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 16711-16724. https://doi.org/10.1109/tcsvt.2026.3713342

Vancouver

Dai D, Li Y, Liu Y, Jia M, Zhang Y, Xia S, et al. FaceCaption-15M: Benchmarking and Enhancing Facial Vision--Language Models. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):16711-16724. doi:10.1109/tcsvt.2026.3713342

BibTeX
@article{Dai2026,
  author = {Dai, Dawei and Li, Yutang and Liu, Yingge and Jia, Mingming and Zhang, Yuanhui and Xia, Shuyin and Wang, Guoyin},
  title = {FaceCaption-15M: Benchmarking and Enhancing Facial Vision--Language Models},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {16711--16724},
  doi = {10.1109/tcsvt.2026.3713342}
}
Kostenlos starten →