SFG-Net: an Efficient Image to Image Translation Approach for Synthetic Fog Generation

Mahmood Hussain Bhat, Trung-Hieu Le, Shih-Chia Huang

Abstract: In this study, we propose a new synthetic fog generation network (SFG-Net), which employs the two-stream image-to-image translation model as a base backbone and integrates a self-attention (SA) module for generating foggy images. With the presence of the SA module between the style pipeline and generator of the proposed SFG-Net, the style structure from style features of the input images is captured to advance the image translation performance. Experimental results show the effectiveness of the proposed SFG-Net in both quantitative evaluations and perceptual quality compared with the competitive method.