Channel Attention GAN Trained With Enhanced Dataset for Single-Image Shadow Removal

Abstract

Even today, where many deep-learning-based methods have been published, single-image shadow removal is a challenging task to achieve high accuracy. This is because the shadow changes depending on various conditions such as the target material or the light source, and it is difficult to estimate all the physical parameters. In this paper, we propose a new single-image shadow removal method (Channel Attention GAN: CANet) using two networks for detecting shadows and removing shadows. Intensity change in shadowed regions has different characteristics depending on the wavelength of light. In addition, the image acquisition system of the camera acquires an image in a state where the RGB values influence each other. Therefore, our method focused on the physical properties of shadows and the camera's image acquisition system. The proposed network has a structure considering the relationship between color channels. When training this network, we modified the color and added some artifacts to the training images in order to make the training dataset more complex. These image processing are based on the shadow model, considering the camera image acquisition system. With these new proposals, our method can remove shadows in all ISTD, ISTD+, SRD, and SRD+ datasets with higher accuracy than the state-of-the-art methods. The code is available on GitHub: https://github.com/ryo-abiko/CANet.

A dynamic alignment and illumination‐aware convolution for shadow removal

“…It is worth mentioning that DAIC has never been trained on Bungalows but still can help the foreground detection method to obtain better results on this dataset, which indicates that our proposed network has certain robustness and practical application value. [9], AEF [13], DHAN [30], CANet [8], and ours, respectively. The experiments demonstrate that DAIC can help existing foreground segmentation methods improve their robustness, and performance improvement can be achieved by adding DAIC without replacing existing foreground segmentation methods.…”

“…We conducted a comprehensive comparison of our proposed method with several state‐of‐the‐art algorithms, including DHAN [30], DC‐ShadowNet [9], SP + M‐Net [11], AEF [13], and CANet [8]. The quantitative evaluations are performed on the ISTD+ and SRD datasets.…”

“…Abiko et al. [8] similarly proposed a stacked GAN model with the addition of a channel attention module. Jin et al.…”

“…There are two types of methods. One is based on generative adversarial network (GAN) [7][8][9][10]. The other is a network model based on physical models [6,[11][12][13][14].…”

“…Using a dynamic field of view enables more accurate feature extraction based on local information to focus the current field of view on the relevant pixel points. In recongnition of the light‐dependent characteristics for the shadow removal task, numerous prior methods have advocated for the generation of shadow masks to encapsulate the light‐dependent characteristics of shadow image [6–8, 16, 17]. To accommodate variations in illumination characteristics, we propose to use Dynamic Weight Module (DWM), to employ dynamic parameter convolution to adapt convolutional operations based on distinct lighting features.…”