Deep neural networks provide excellent performance for inverse problems such as denoising. However, neural networks can be sensitive to adversarial or worst-case perturbations. This raises the question of whether such networks can be trained efficiently to be worst-case robust.

In this supplemental material, we first present the detailed network architecture and parameters of the proposed approach in Sec. A. We further provide more analysis of the proposed method and ablation studies in Sec. B. Section C shows some qualitative results for potential applications of the proposed approach on medical imaging and imaging in astronomy. Our goal of this section is to complement the results in the main paper (in particular Tab. 3 in the The basic reconstruction network that contains three residual blocks followed by one convolutional layer is denoted as Basic reconstruction . We thus disable our proposed feature refinement network for all the baseline methods in this section. The results in Tab. 8 demonstrate that the deep features are more effective for extracting useful The last row in Tab. 8 reports the results of the proposed approach with the deep Wiener deconvolution B.2 Effectiveness of learned deep features with a basic reconstruction network In the main paper, we compare the proposed method with state-of-the-art methods on the dataset of Levin et al. Figure 6: Illustration of learned deep features.

We present a simple and effective approach for non-blind image deblurring, combining classical techniques and deep learning.

Why it makes sense to deblur the extracted features? Violations can be successfully compensated by the feature refinement. We will discuss these in detail and add corresponding results in the revised paper. The reason why the improvement in Tab. 3 is not so Our PSNR results are 25.57 Tabs. 3 and 5 are evaluated on [19] and [16], respectively, with The contributions are summarized in L50-61.

U-Net and other U-shaped architectures have achieved significant success in image deconvolution tasks. However, challenges have emerged, as these methods might generate unrealistic artifacts or hallucinations, which can interfere with analysis in safety-critical scenarios. This paper introduces a novel approach for quantifying and comprehending hallucination artifacts to ensure trustworthy computer vision models. Our method, termed the Conformal Hallucination Estimation Metric (CHEM), is applicable to any image reconstruction model, enabling efficient identification and quantification of hallucination artifacts. It offers two key advantages: it leverages wavelet and shearlet representations to efficiently extract hallucinations of image features and uses conformalized quantile regression to assess hallucination levels in a distribution-free manner . Furthermore, from an approximation theoretical perspective, we explore the reasons why U-shaped networks are prone to hallucinations. W e test the proposed approach on the CANDELS astronomical image dataset with models such as U-Net, Swin-UNet, and Learnlets, and provide new perspectives on hallucination from different aspects in deep learning-based image processing.

Fluorescent calcium indicators are a popular means for observing the spiking activity of large neuronal populations.

In this paper we introduce a natural image prior that directly represents a Gaussian-smoothed version of the natural image distribution.

Neural Information Processing Systems http://nips.cc/

Neural Information Processing Systems http://nips.cc/

This is an idealization of the usual way to train neural networks with a large hidden layer.