Progressive downsampling and adaptive guidance networks for dynamic scene deblurring

Highlights：

• We propose a novel progressive downsampling and adaptive guidance network for retaining more the strong edges and other high-frequency information of the blurry images, so as to make the network model learn a more effective dynamic scene deblurring mapping.

• In the proposed network, we design a multiscale blended activation residual block to learn the nonlinear characteristics of dynamic scene blur, which can alleviate the performance saturation problem caused by a single activation function and improve multiscale feature extraction ability.

• We propose a multisupervision strategy for making the proposed network learn more robust and effective features and making the network possess more stable training and faster convergence.