SCAN: Spatial and Channel Attention Normalization for Image Inpainting

Image inpainting focuses on predicting contents with shape structure and consistent details in damaged regions. Recent approaches based on convolutional neural network (CNN) have shown promising results via adversarial learning, attention mechanism, and various loss functions. This paper introduces a novel module named Spatial and Channel Attention Normalization (SCAN), combining attention mechanisms in spatial and channel dimension and normalization to handle complex information of known regions while avoiding its misuse. Experiments on the varies datasets indicate that the performance of the proposed method outperforms the current state-of-the-art (SOTA) inpainting approaches.