Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions (1711.08141v2)

Published 22 Nov 2017 in cs.CV

Abstract: Neural networks rely on convolutions to aggregate spatial information. However, spatial convolutions are expensive in terms of model size and computation, both of which grow quadratically with respect to kernel size. In this paper, we present a parameter-free, FLOP-free "shift" operation as an alternative to spatial convolutions. We fuse shifts and point-wise convolutions to construct end-to-end trainable shift-based modules, with a hyperparameter characterizing the tradeoff between accuracy and efficiency. To demonstrate the operation's efficacy, we replace ResNet's 3x3 convolutions with shift-based modules for improved CIFAR10 and CIFAR100 accuracy using 60% fewer parameters; we additionally demonstrate the operation's resilience to parameter reduction on ImageNet, outperforming ResNet family members. We finally show the shift operation's applicability across domains, achieving strong performance with fewer parameters on classification, face verification and style transfer.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (9)
  1. Bichen Wu (52 papers)
  2. Alvin Wan (16 papers)
  3. Xiangyu Yue (93 papers)
  4. Peter Jin (9 papers)
  5. Sicheng Zhao (53 papers)
  6. Noah Golmant (2 papers)
  7. Amir Gholaminejad (1 paper)
  8. Joseph Gonzalez (35 papers)
  9. Kurt Keutzer (200 papers)
Citations (345)

Summary

Overview of LaTeX Author Guidelines for CVPR Proceedings

The paper "LaTeX Author Guidelines for CVPR Proceedings" serves as an essential manual for authors intending to submit their manuscripts for the Computer Vision and Pattern Recognition (CVPR) conference proceedings. The paper meticulously outlines necessary formatting, submission, and composition guidelines, ensuring uniformity and adherence to standards across submissions.

Structural and Formatting Instructions

The paper emphasizes stringent formatting rules to maintain consistency. Authors are instructed to adhere to a two-column layout with specific dimensions for text width and spacing, ensuring that each submission's presentation is uniform. The manuscript's primary sections, including the abstract, main text, and references, must follow prescribed type-styles and font sizes: for instance, Times 10-point for the main text and Times 12-point boldface for the first-order headings.

Additionally, the document insists on using a printed ruler in the review version to aid reviewers in referencing specific sections or lines. The presence of such guidelines is crucial in maintaining the quality and standard expected from a high-profile conference like CVPR.

Content and Submission Guidelines

The paper outlines significant policies related to content creation and submission, such as dual submission protocols, language necessities, and page length restrictions. The dual submission section, for example, underscores the imperative nature of consulting CVPR's latest web page guidelines to avoid conflicts, thus preventing potential ethical issues and ensuring only novel, unpublished work undergoes peer review.

The guidelines specifically caution against overlength submissions by setting the maximum at eight pages, excluding references. Such stipulations serve as preventive measures against the arbitrary extension of content, potentially facilitating a more efficient review process by maintaining a manageable scope for reviewers.

Blind Review and Citation Protocol

An interesting aspect of the paper is the clarifications surrounding the blind review process, which mandates the removal of explicit author identification from the manuscript yet allows citations to the authors’ prior works. Authors are advised to use third-person language when citing their previous works, which maintains anonymity while providing context to the research presented. This section is crucial for ensuring impartiality and preventing bias during the peer review.

Furthermore, the paper provides precise instructions regarding citation styles—advocating for numerical order rather than chronological in multi-citation instances. This standardizes referencing and enhances the ease of navigating academic discourse.

Implications and Future Directions

The structured approach prescribed by the guidelines not only serves practical purposes in manuscript preparation but also has broader implications on academic publishing. By enforcing strict adherence to format and review protocols, the guidelines help to maintain a high quality of submissions and uphold the integrity of the peer review process. Such rigor is essential for maintaining CVPR's esteemed reputation in the domain of computer vision research.

As the field of AI and computer vision continues to evolve, these guidelines may undergo further modifications to incorporate new standards in paper formatting and content, reflecting technological advancements and community needs. The adaptability of these guidelines will be instrumental in continuing CVPR's role as a leading venue for groundbreaking research in computer vision.