Zhang, Lin, Zhao, Huihuang, Meng, Weiliang, Yang, Yuan, Rosin, Paul L.  ORCID: https://orcid.org/0000-0002-4965-3884, Lai, Yu-Kun ORCID: https://orcid.org/0000-0002-2094-5680 and Wang, Yaonan
2026.
FaceEditor: Text-driven and mask-constrained face attribute editing.
Pattern Recognition
179
(Part A)
, 113559.
10.1016/j.patcog.2026.113559
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (3MB) | Preview |
Abstract
Face image editing has advanced in recent years, with most methods using multimodal conditional guidance to achieve realistic results. However, these methods cannot intuitively edit specific face image regions, and entangled semantics make preserving unrelated attributes difficult. To address these problems, this paper proposes a unified image manipulation framework named FaceEditor, which supports both text and masks for individually or jointly editing facial attributes. The key idea is to train a coarse-to-fine Editing Direction Mapper (ED Mapper) to predict latent manipulation directions from text. During inference, segmentation masks constrain the blending of latent codes and editing features in the feature space, enabling localized and controllable image editing. Additionally, we designed a Global Modulation Module (GMM) to globally blend and optimize latent features at different levels, further enhancing the model’s disentangled manipulation capabilities and editing precision. Experimental results show that FaceEditor outperforms existing methods in accuracy, visual realism, and preservation of unrelated attributes, while enabling real-time text-only editing at 0.61s per image. Code is available at https://github.com/Zlin0530/FaceEditor.
| Item Type: | Article |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Computer Science & Informatics |
| Additional Information: | RRS policy applied |
| Publisher: | Elsevier BV |
| ISSN: | 0031-3203 |
| Date of First Compliant Deposit: | 16 April 2026 |
| Date of Acceptance: | 20 March 2026 |
| Last Modified: | 16 Apr 2026 09:00 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/186087 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions