Wanglong Lu

Wanglong Lu

Nasdaq

GRIG: Data-Efficient Generative Residual Image Inpainting

Computational Visual Media 11(6), 1329-1361, 2025 · First author

Wanglong Lu, Xianta Jiang, Xiaogang Jin, Yong-Liang Yang, Minglun Gong, Kaijie Shi, Tao Wang, Hanli Zhao

Watch demo

framework

Abstract:

GRIG addresses image inpainting when training data are limited. Its iterative residual reasoning combines CNN feature extraction with Transformer-based global reasoning inside a generative adversarial framework. Image-level and patch-level discriminators, together with forged-patch adversarial training, improve texture fidelity and fine detail. The published study evaluates data-efficient inpainting against previous methods, using quantitative comparisons and visual results.

[paper] [doi] [published pdf] [arxiv] [homepage]

Recommended citation:

@article{lu2025grig,
  title = {GRIG: Data-Efficient Generative Residual Image Inpainting},
  author = {Lu, Wanglong and Jiang, Xianta and Jin, Xiaogang and Yang, Yong-Liang and Gong, Minglun and Shi, Kaijie and Wang, Tao and Zhao, Hanli},
  journal = {Computational Visual Media},
  volume = {11},
  number = {6},
  pages = {1329--1361},
  year = {2025},
  doi = {10.26599/CVM.2025.9450408}
}