A Multistage with Multiattention Network for Single Image Dehazing

Figure 4

Encoder-decoder subnetwork is based on U-Net, and down-sampling layer is used to reduce the computation. The down-sampling and up-sampling are implemented by a 3 × 3 convolution.