T-Net: Deep Stacked Scale-Iteration Network for Image Dehazing

Haze reduces the visibility of image content and leads to failure in handling subsequent computer vision tasks. In this paper, we address problem single dehazing by proposing a network named T-Net, which consists backbone based on U-Net architecture dual attention module. Multi-scale feature fusion can be achieved using skip connections with new strategy. Furthermore, repeatedly unfolding plain Stack T-Net is proposed take advantage dependence deep features across stages via recursive To reduce parameters, intra-stage computation ResNet adopted our T-Net. We both stage-wise result original hazy as input each finally output prediction clean image. Experimental results synthetic real-world images demonstrate that advanced perform favorably against state-of-the-art algorithms show could further improve effect, demonstrating effectiveness