SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation

Xian, Yukai; Lee, Yurui; Yan, Liang; Shen, Te; Lan, Ping; Zhao, Qijun; Zhang, Yi

doi:10.3390/sym17101643

This is an early access version, the complete PDF, HTML, and XML versions will be available soon.

Open AccessArticle

SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation

by

Yukai Xian

^1,†

,

Yurui Lee

^2,†,

Liang Yan

^2,*,

Te Shen

¹,

Ping Lan

¹

,

Qijun Zhao

^1,3,* and

Yi Zhang

⁴

¹

School of Information Science and Technology, Tibet University, Lhasa 850000, China

²

School of Humanities, Tibet University, Lhasa 850000, China

³

College of Computer Science, Sichuan University, Chengdu 610065, China

⁴

School of Business and Media, Lanzhou University of Finance and Economics, Lanzhou 730020, China

^*

Authors to whom correspondence should be addressed.

^†

These authors contributed equally to this work.

Symmetry 2025, 17(10), 1643; https://doi.org/10.3390/sym17101643

Submission received: 21 August 2025 / Revised: 26 September 2025 / Accepted: 2 October 2025 / Published: 3 October 2025

(This article belongs to the Special Issue Symmetry/Asymmetry in Image Processing and Computer Vision)

Download Versions Notes

Abstract

Thangka paintings, as intricate forms of Tibetan Buddhist art, present unique challenges for image segmentation due to their densely arranged symbolic elements, complex color patterns, and strong structural symmetry. To address these difficulties, we propose SPIRIT, a structure-aware and prompt-guided diffusion segmentation framework tailored for Thangka images. Our method incorporates a support-query-encoding scheme to exploit limited labeled samples and introduces semantic guided attention fusion to integrate symbolic knowledge into the denoising process. Moreover, we design a symmetry-aware refinement module to explicitly preserve bilateral and radial symmetries, enhancing both accuracy and interpretability. Experimental results on our curated Thangka dataset and the artistic ArtBench benchmark demonstrate that our approach achieves 88.3% mIoU on Thangka and 86.1% mIoU on ArtBench, outperforming the strongest baseline by 6.1% and 5.6% mIoU, respectively. These results confirm that SPIRIT not only captures fine-grained details, but also excels in segmenting structurally complex regions of artistic imagery.

Keywords: Thangka image segmentation; diffusion models; symmetry-aware segmentation; cultural heritage preservation; prompt-guided learning

Share and Cite

MDPI and ACS Style

Xian, Y.; Lee, Y.; Yan, L.; Shen, T.; Lan, P.; Zhao, Q.; Zhang, Y. SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation. Symmetry 2025, 17, 1643. https://doi.org/10.3390/sym17101643

AMA Style

Xian Y, Lee Y, Yan L, Shen T, Lan P, Zhao Q, Zhang Y. SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation. Symmetry. 2025; 17(10):1643. https://doi.org/10.3390/sym17101643

Chicago/Turabian Style

Xian, Yukai, Yurui Lee, Liang Yan, Te Shen, Ping Lan, Qijun Zhao, and Yi Zhang. 2025. "SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation" Symmetry 17, no. 10: 1643. https://doi.org/10.3390/sym17101643

APA Style

Xian, Y., Lee, Y., Yan, L., Shen, T., Lan, P., Zhao, Q., & Zhang, Y. (2025). SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation. Symmetry, 17(10), 1643. https://doi.org/10.3390/sym17101643

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu

SPIRIT: Symmetry-Prior Informed Diffusion for Thangka Segmentation

Abstract

Share and Cite

Article Metrics

Article Access Statistics

Further Information

Guidelines

MDPI Initiatives

Follow MDPI