| Autor: |
Cao, Zhiyi, Xie, Bin, Huo, Lina, Shao, Mingwen |
| Zdroj: |
International Journal of Machine Learning & Cybernetics; Jun2024, Vol. 15 Issue 6, p2481-2493, 13p |
| Abstrakt: |
Due to advances in adversarial learning and self-supervised learning, global mosaic removal no longer requires paired samples to achieve superior results. However, most existing methods require a large number of samples to capture the distribution of mosaic characteristics, which severely limits their scope of application. To solve this problem, this paper proposes a conditional contrastive learning model for unpaired global mosaic removal with a few samples. Firstly, a conditional contrastive learning loss function is proposed to eliminate inappropriate negative samples in current contrastive learning methods and improve unpaired global mosaic removal results. Then, in addition to conditional contrastive learning loss, a global contrastive loss function is proposed to obtain global features from all layers, and a local contrastive loss function is proposed to obtain correct block-level feature representations. Finally, they work together to obtain the best unpaired global mosaic removal results with a few samples. Experiments on three datasets show that the proposed model has achieved leading results in both quantitative and qualitative analysis of three state-of-the-art models for global mosaic removal. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink