SimMatch: Semi-supervised Learning with Similarity Matching (2203.06915v2)

Abstract: Learning with few labeled data has been a longstanding problem in the computer vision and machine learning research community. In this paper, we introduced a new semi-supervised learning framework, SimMatch, which simultaneously considers semantic similarity and instance similarity. In SimMatch, the consistency regularization will be applied on both semantic-level and instance-level. The different augmented views of the same instance are encouraged to have the same class prediction and similar similarity relationship respected to other instances. Next, we instantiated a labeled memory buffer to fully leverage the ground truth labels on instance-level and bridge the gaps between the semantic and instance similarities. Finally, we proposed the \textit{unfolding} and \textit{aggregation} operation which allows these two similarities be isomorphically transformed with each other. In this way, the semantic and instance pseudo-labels can be mutually propagated to generate more high-quality and reliable matching targets. Extensive experimental results demonstrate that SimMatch improves the performance of semi-supervised learning tasks across different benchmark datasets and different settings. Notably, with 400 epochs of training, SimMatch achieves 67.2\%, and 74.4\% Top-1 Accuracy with 1\% and 10\% labeled examples on ImageNet, which significantly outperforms the baseline methods and is better than previous semi-supervised learning frameworks. Code and pre-trained models are available at https://github.com/KyleZheng1997/simmatch.

Insights from the CVPR Proceedings Author Guidelines

Introduction to the Guidelines

The paper presents a comprehensive set of instructions for authors preparing manuscripts for the IEEE Computer Society Press, specifically for submission to the Conference on Computer Vision and Pattern Recognition (CVPR). This revised guide encompasses several updates from previous versions, notably removing the caveat against using adhesive tape for attaching artwork. The guidelines cover a broad range of topics from language requirements and dual submission policy to formatting specifics such as paper length, the inclusion of a ruler for review purposes, and recommendations for mathematics notation.

Manuscript Preparation and Submission Details

Language and Dual Submission

• Manuscripts must be written in English.
• The guidelines reiterate CVPR's policy on dual submissions, directing authors to specific information on the CVPR 2022 webpage.

Paper Length and Formatting

• The maximum length for papers, excluding references, is set at eight pages.
• There are no extra page charges for CVPR 2022, and overlength papers will not be reviewed.
• A printed ruler is defined for inclusion in the review version to facilitate specific feedback from reviewers.

Identifying Information for Blind Review

• The paper dispels common misconceptions about the anonymization process for blind review, clarifying that self-citation is allowed as long as it is done in the third person to prevent author identification.

Formatting Specifics

General Layout

• The text must adhere to a two-column format within specified dimensions and margins, with detailed instructions on the main title, author names, and affiliations presentation.
• Type style and font guidelines suggest using Times or the closest available font, with specific size and style details for different text elements.

Figures, Tables, and Mathematics

• Instructions for figure and table presentation emphasize centering and compatibility in print format, suggesting specific caption styles and the use of standard LaTeX commands for consistent formatting.
• For mathematical notation, the paper underscores the importance of numbering all equations for easy reference and suggests consulting Mermin's guide for writing mathematics.

Ethical Considerations and Final Submission

The guidelines address ethical considerations including the handling of supplemental materials and parallel submissions to maintain the integrity of the double-blind review process. In the final stages of manuscript preparation, authors are reminded to include a signed IEEE copyright release form and to remove any page numbers or the review ruler from the camera-ready copy.

Implications and Future Directions

The clear, detailed guidelines provided in the paper not only streamline the submission process for CVPR but also have broader implications for the standardization of manuscript preparation across conferences in the computer vision and pattern recognition community. By elucidating common pitfalls, especially in the blind review process, the guidelines contribute to a fairer, more efficient review process.

Future evolutions of these guidelines may further adapt to technological advancements and changing ethical standards in scientific publishing. One can anticipate that future revisions might address emerging challenges such as the citation of preprints, the role of AI in generating research artifacts, and more nuanced aspects of accessibility and inclusivity in visual material presentation.

In conclusion, the CVPR proceedings author guidelines provide an essential resource for authors, ensuring consistency, fairness, and high quality in conference submissions. By adhering to these guidelines, authors contribute to the advancement and integrity of research in the computer vision and pattern recognition fields.