Fig. 1. Overall framework of PADNet (magnified the pixel values of the polarization degree images by a factor of 10)

Fig. 2. Structural frameworks of the traditional U-shaped network and PADNet. (a) Structure of the U-shaped network; (b) dual-encoding structure of PADNet

Fig. 3. Structure diagram of dynamic large kernel convolution module

Fig. 4. Structure diagram of adaptive dynamic feature fusion module

Fig. 5. Collection scheme of IHP dataset. (a) Controllable environmental data acquisition system; (b) examples of collected data

Fig. 6. Visualization experiment results on the IHP dataset

Fig. 7. Visualization experiment results on the Cityscapes-DBF dataset are also presented

Fig. 8. Qualitative comparison results of dehazing effects in real outdoor scenes using different ethods

Fig. 9. Visualization comparison results of different methods in ablation studies, the numerical values in the lower right corner represent the quantitative results evaluated by PSNR/SSIM