We present a simple and effective approach for non-blind image deblurring, combining classical techniques and deep learning. In contrast to existing methods that deblur the image directly in the standard image space, we propose to perform an explicit deconvolution process in a feature space by integrating a classical Wiener deconvolution framework with learned deep features. A multi-scale feature refinement module then predicts the deblurred image from the deconvolved deep features, progressively recovering detail and small-scale structures. The proposed model is trained in an end-to-end manner and evaluated on scenarios with both simulated and real-world image blur. Our extensive experimental results show that the proposed deep Wiener deconvolution network facilitates deblurred results with visibly fewer artifacts. Moreover, our approach quantitatively outperforms state-of-the-art non-blind image deblurring methods by a wide margin.

Clarity: Overall, the paper is written well, and easy to understand the main idea. However, there are some unclear points as follows: 1) There is no explanation why the feature-based Wiener deconvolution is better than image-based method. For example, is the proposed feature-based method better for estimating more accurate signal and noise level? Is the model "Ours w/o Wiener" guided by the input blur kernel? If you guide the kernel, how do you guide it to the network?

Overall, the reviewers were positive about the paper: the experimental results are very good, and the idea of using Wiener deconvolution in the feature space is interesting. After the rebuttal and discussion, the reviewers unanimously voted for acceptance. Please put the clarifications in the rebuttal into the final version.

We present a simple and effective approach for non-blind image deblurring, combining classical techniques and deep learning. In contrast to existing methods that deblur the image directly in the standard image space, we propose to perform an explicit deconvolution process in a feature space by integrating a classical Wiener deconvolution framework with learned deep features. A multi-scale feature refinement module then predicts the deblurred image from the deconvolved deep features, progressively recovering detail and small-scale structures. The proposed model is trained in an end-to-end manner and evaluated on scenarios with both simulated and real-world image blur. Our extensive experimental results show that the proposed deep Wiener deconvolution network facilitates deblurred results with visibly fewer artifacts. Moreover, our approach quantitatively outperforms state-of-the-art non-blind image deblurring methods by a wide margin.