A Deep Motion Deblurring Network Using Channel Adaptive Residual Module

In this paper, we solve the problem of dynamic scenes deblurring with motion blur. Restoration images in presence blur necessitates a network design that receptive field can completely cover all areas need to be deblurred, while existing increases by continuously stacking ordinary convolutional layer or increasing size convolution kernel. However, these methods inevitably increase computational burden network. We propose novel architecture consisting channel adaptive residual module. Different features blurred image are extracted and distributed on each feature channel. Our calculate weight through learning, extract adaptively according different degrees blurring importance information. embed module modified encoder-decoder skip connections achieve multi-scale fusion for further performance improvement. The extensive comparison techniques baseline scene dataset shows proposed effectively realize deblurring, accuracy speed comparable techniques.