收藏切换
Research on depth estimation and portrait segmentation based on diffusion models
收藏切换
PDF
Zongbo DONG, Yifan WANG, Lijun WANG*, Huchuan LU
Science & Technology Review | 2025, 43(22) : 98 - 107
Less
收藏切换
Science & Technology Review | 2025, 43(22): 98-107
Papers
Research on depth estimation and portrait segmentation based on diffusion models
Full
Zongbo DONG, Yifan WANG, Lijun WANG*, Huchuan LU
Affiliations
  • Dalian University of Technology School of Future Technology, Dalian 116024, China
Published: 2025-11-28 doi: 10.3981/j.issn.1000-7857.2025.05.00059
Outline
收藏切换

While diffusion models have demonstrated remarkable capabilities in generative tasks, their application to visual perception tasks such as depth estimation and portrait segmentation remains underexplored. This paper proposes Diffusion Perception, a unified framework based on diffusion models for high−quality depth estimation and portrait segmentation. By reformulating traditional perception tasks as conditional generation problems, the framework leverages the denoising characteristics of latent diffusion models (LDMs) to optimize prediction results in latent space. The innovative design incorporates three core processing stages: multimodal feature encoding, noise input prediction, and text−controlled feature extraction and reconstruction, enabling the transition of diffusion models from generative paradigms to visual perception task paradigms. Experimental results demonstrate that on our custom depth estimation dataset, the proposed method achieves evaluation metrics of 93.98% Relative Accuracy (RR), 99.61% Plane Estimation Accuracy (Plane), and 93.61% Scene Consistency (Consistence), outperforming existing state−of−the−art depth estimation methods. Furthermore, in portrait segmentation tasks, the method achieves Intersection over Union (IoU) and mean IoU (mIoU) scores of 96.98% and 91.98% respectively, surpassing existing segmentation algorithms. This study provides novel insights into applying diffusion models in visual perception, where their generative paradigm naturally handles prediction uncertainty and is well−suited for robust perception in dynamic environments.

diffusion models  /  depth estimation  /  portrait segmentation  /  fully convolutional networks  /  deep learning
Zongbo DONG, Yifan WANG, Lijun WANG, Huchuan LU. Research on depth estimation and portrait segmentation based on diffusion models[J]. Science & Technology Review, 2025 , 43 (22) : 98 -107 . DOI: 10.3981/j.issn.1000-7857.2025.05.00059
Year 2025 volume 43 Issue 22
PDF
1199
674
Cite this Article
BibTeX
Article Info
doi: 10.3981/j.issn.1000-7857.2025.05.00059
  • Receive Date:2025-05-12
  • Online Date:2025-12-29
  • Published:2025-11-28
Article Data
Affiliations
History
  • Received:2025-05-12
  • Revised:2025-11-03
Funding
Affiliations
    Dalian University of Technology School of Future Technology, Dalian 116024, China
References
Share
https://castjournals.cast.org.cn/joweb/kjdb/EN/10.3981/j.issn.1000-7857.2025.05.00059
Share to
QR

Scan QR to access full text

Cite this article
BibTeX
Citations
表12种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏
  • BibTeX
  • EndNote
  • RefWorks
  • TxT