Soft–Hard Attention U-Net Model and Benchmark Dataset for Multiscale Image Shadow Removal

Effective shadow removal is pivotal in enhancing visual image quality in applications such as computer vision and digital photography. During the last decades physics-based and machine learning methodologies have been proposed; however, most of them have limited capacity in capturing complex shadow patterns due to restrictive model assumptions, neglecting the fact that shadows usually appear at different scales. Existing benchmarking shadow removal datasets have a limited number of images with simple scenes containing uniform shadows cast by single objects, whereas only a few of them include both manual shadow annotations and paired shadow-free images. To address these limitations, in natural and urban scenes with complex shadows, the contribution of this study is twofold, it: a) proposes a novel deep learning architecture, named Soft-Hard Attention U-net (SHAU), focusing on multiscale shadow removal; b) provides a novel synthetic dataset, named Multiscale Shadow Removal Dataset (MSRD), containing complex shadow patterns of multiple scales, serving as a privacy-preserving benchmark dataset. SHAU incorporates both soft and hard attention modules, along with multiscale feature extraction blocks, enable effective shadow removal of different scales. Experimental results show that SHAU outperforms state-of-the-art methods, improving Peak Signal-to-Noise Ratio and Root Mean Square Error in shadow areas by 18.7% and 55.1%, respectively.

# Electrical engineering. Electronics. Nuclear engineering # Shadow removal # multiscale shadows # convolutional neural network # U-Net # synthetic dataset

Eirini Cholopoulou, Dimitrios E. Diamantis, Dimitra-Christina C. Koutsiou, et al. (2026). "Soft–Hard Attention U-Net Model and Benchmark Dataset for Multiscale Image Shadow Removal". IEEE Access. https://doi.org/10.1109/access.2026.3728048