Guided Depth Super-Resolution by Deep Anisotropic Diffusion

Abstract: Performing super-resolution of a depth image using the guidance from an RGB image is a problem that concerns several fields, such as robotics, medical imaging, and remote sensing. While deep learning methods have achieved good results in this problem, recent work highlighted the value of combining modern methods with more formal frameworks. In this work, we propose a novel approach which combines guided anisotropic diffusion with a deep convolutional network and advances the state of the art for guided depth super-resolution. The edge transferring/enhancing properties of the diffusion are boosted by the contextual reasoning capabilities of modern networks, and a strict adjustment step guarantees perfect adherence to the source image. We achieve unprecedented results in three commonly used benchmarks for guided depth super-resolution. The performance gain compared to other methods is the largest at larger scales, such as x32 scaling. Code (https://github.com/prs-eth/Diffusion-Super-Resolution) for the proposed method is available to promote reproducibility of our results.

• The paper introduces a novel deep anisotropic diffusion framework to perform guided depth super-resolution.
• It integrates learned diffusion coefficients with a convolutional network to maintain edge details and suppress noise.
• Experimental results demonstrate significant improvements over traditional methods in both qualitative and quantitative metrics.

Overview of CVPR Proceedings Author Guidelines

The document, titled "LaTeX Author Guidelines for CVPR Proceedings," serves as a comprehensive manual for authors intending to submit manuscripts to the Conference on Computer Vision and Pattern Recognition (CVPR). The paper emphasizes formatting standards and procedural protocols necessary for the preparation of submissions using LaTeX, a prevalent document preparation system in academia.

The guidelines articulated in the paper underscore several pivotal elements required to ensure consistency across submissions. Firstly, it establishes the importance of English as the primary language of the manuscript, which aligns with the global nature of the conference. The policy on dual submission is addressed, highlighting ethical concerns and guiding authors through intellectual property management when submitting to multiple venues.

Regarding the structural aspects of the paper, the authors have delineated restrictions on paper length, distinctly pointing out that the total length excluding references should not exceed eight pages. There is a clear directive that any deviations in margins and formatting aimed to circumvent this limitation will result in the paper not being reviewed. This stipulation underscores the conference's approach to maintain a level playing field for all submissions.

An interesting feature of the style guide is its inclusion of the 'ruler,' which enables reviewers to precisely reference sections in the document. Although not expected in the final submission, its presence during review stages is pivotal in facilitating clear communication between authors and reviewers.

For mathematical notation, the guidelines insist on proper numbering of sections and equations to ensure readability and ease of reference by future readers. The paper also provides prudent advice for maintaining anonymity during blind review by recommending specific linguistic approaches for citing prior work. This reflects an adherence to fair reviewing practices.

The document is exhaustive in detailing formatting aspects such as type style and fonts, footnotes, cross-references, and page layout, all crafted to create a uniform appearance among contributions and ease of readability.

On practical elements, the guidelines touch upon the use of color in visuals, bearing in mind accessibility concerns such as color vision deficiencies among readers and reviewers. This attention to detail extends to recommendations for figure placement and scaling within LaTeX, as visual clarity remains crucial for comprehension.

Finally, the paper stipulates the requirement to include a signed IEEE copyright release form upon submission of the finished paper. This legal aspect serves to protect both authors and conference organizers regarding the dissemination of the work.

In conclusion, the CVPR author guidelines document exemplifies a robust structure designed to facilitate the creation of scientifically rigorous and formally coherent contributions to one of the leading conferences in computer vision. The minutiae captured in this paper not only support authors in compliant manuscript preparation but also reveal the evolving nature of conference publication standards, reflecting broader trends in scholarly communication and technological advancement. These guidelines will continue to serve future iterations, enabling the CVPR to maintain its influential position in the field.