Abstract
Rainy days usually degrade the visual effect of images and videos. At present, most deraining models for single images adopt gradual optimization or elimination to remove rain streaks, but actually with relatively low efficiency in real tasks. An efficient one-stage deraining model, Efficient Transformer Derain Network (ETDNet), is proposed to remove rain streaks in single images efficiently. A new Transformer architecture is designed to provide rich multiple scales and context information, making the model extract features in a coarse-to-fine way. Multiple expansion filters with different expansion rates are embedded to predict the suitable kernel for each pixel of the rainy image in a multi-scale way. A multi-scale Loss Function is introduced to restore features with high-fidelity and detail textures. Experiments on Rain100L, Rain100H, and SPA datasets show that compared with other models, the proposed ETDNet reaches the highest PSNR and SSIM values at the fastest speed.
Highlights
R AINY days will decrease visibility, and the dense rainwater will cause diffuse reflection, making it hard to restore images with details on rainy days
In this paper, a one-stage deraining model for images based on deep learning is introduced from a new perspective
Compared with other deraining models, the Efficient Transformer Derain Network (ETDNet) model proposed in this paper can remove rain streaks faster and recover more texture details
Summary
R AINY days will decrease visibility, and the dense rainwater will cause diffuse reflection, making it hard to restore images with details on rainy days. Representative methods include RCDNet [9], JORDER [10], SNet [11], MPRNet [12], etc They first predict the unknown rain layers and background images to be recovered and use various network structures to normalize and separate them. Those methods often involve many iterations and continuous optimization, which cannot cover all rainfall conditions and require substantial computational overhead. The detail texture restored by those methods will be too smooth as the rain streaks and the background are essentially overlapping in feature space, resulting in losing details in the non-rainfall area of the restored background image.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
