Rethinking Coarse-to-Fine Approach in Single Image Deblurring (2108.05054v2)

Abstract: Coarse-to-fine strategies have been extensively used for the architecture design of single image deblurring networks. Conventional methods typically stack sub-networks with multi-scale input images and gradually improve sharpness of images from the bottom sub-network to the top sub-network, yielding inevitably high computational costs. Toward a fast and accurate deblurring network design, we revisit the coarse-to-fine strategy and present a multi-input multi-output U-net (MIMO-UNet). The MIMO-UNet has three distinct features. First, the single encoder of the MIMO-UNet takes multi-scale input images to ease the difficulty of training. Second, the single decoder of the MIMO-UNet outputs multiple deblurred images with different scales to mimic multi-cascaded U-nets using a single U-shaped network. Last, asymmetric feature fusion is introduced to merge multi-scale features in an efficient manner. Extensive experiments on the GoPro and RealBlur datasets demonstrate that the proposed network outperforms the state-of-the-art methods in terms of both accuracy and computational complexity. Source code is available for research purposes at https://github.com/chosj95/MIMO-UNet.

• The paper introduces an innovative coarse-to-fine framework that improves deblurring performance over traditional methods.
• It employs a multi-scale architecture to progressively refine image details and effectively reduce blur artifacts.
• Experimental results demonstrate significant enhancements in visual quality and quantitative metrics, offering promising real-world applications.

Overview of LaTeX Author Guidelines for ICCV Proceedings

The document under review provides a comprehensive guide intended for authors submitting manuscripts to the International Conference on Computer Vision (ICCV). It meticulously outlines the necessary formatting and submission protocols, as necessitated by the IEEE Computer Society Press. The paper does not present new research findings; rather, it serves as a structural blueprint for manuscript preparation.

Structural and Formatting Directives

Key sections of the guidelines are methodically divided, emphasizing aspects such as language, submission policies, and formatting standards. Authors are reminded to compose their manuscripts in English and to be cognizant of the dual submission policy, reinforcing adherence to ethical research practices.

• Paper Length: Manuscripts should not exceed eight pages excluding references. This strict page limit is emphasized to streamline the review process, ensuring that all submissions are evaluated uniformly.
• Review Process: A defining feature is the extensive details on preparing documents for blind review. Emphasis is placed on anonymizing self-references without omitting crucial citations, fostering an unbiased review environment.
• Formatting Specifics: Authors must adhere to two-column text layouts, stipulating precise measurements for margins and font styles. A comprehensive breakdown of type-styles for various elements such as titles, section headings, and captions is provided.

The document also prescribes the inclusion of a "ruler" for preliminary submissions, facilitating targeted feedback from reviewers. However, this element is to be excluded from the final camera-ready copy.

Visual Elements and Mathematical Notations

The guidelines suggest utilizing figures and equations effectively, with directives for sizing and label placements to ensure clarity in print formats. Notably, sectioning for mathematics involves mandatory equation numbering, enhancing reference accuracy for subsequent readers. Authors are advised on integrating visuals and graphs such that they retain their efficacy when printed.

References and Citation Styles

Authors are directed to list references in a numerical sequence, adopting a consistent citation format throughout the manuscript. This systematic approach underscores the necessity for scholarly cohesion and integrity.

Practical Implications and Future Considerations

Although primarily procedural, adherence to these guidelines influences the broader academic community by upholding the rigor and clarity expected of conference submissions. Ensuring conformity with these standards not only facilitates a smoother review process but also enhances the dissemination of research findings to the computer vision research community.

Future iterations of these guidelines may evolve to accommodate advancements in document preparation technologies and changing publication norms. Authors should remain informed of updates to these guidelines to ensure ongoing compliance and presentation quality.

In conclusion, this document serves as an essential reference for ICCV authors, meticulously detailing the formatting and ethical standards required to present research effectively. The rigorous enforcement of these protocols encourages high-quality submissions that match the standards of the IEEE Computer Society and the larger academic community within the computer vision domain.