Self-Support Few-Shot Semantic Segmentation (2207.11549v1)

Abstract: Existing few-shot segmentation methods have achieved great progress based on the support-query matching framework. But they still heavily suffer from the limited coverage of intra-class variations from the few-shot supports provided. Motivated by the simple Gestalt principle that pixels belonging to the same object are more similar than those to different objects of same class, we propose a novel self-support matching strategy to alleviate this problem, which uses query prototypes to match query features, where the query prototypes are collected from high-confidence query predictions. This strategy can effectively capture the consistent underlying characteristics of the query objects, and thus fittingly match query features. We also propose an adaptive self-support background prototype generation module and self-support loss to further facilitate the self-support matching procedure. Our self-support network substantially improves the prototype quality, benefits more improvement from stronger backbones and more supports, and achieves SOTA on multiple datasets. Codes are at \url{https://github.com/fanq15/SSP}.

• The paper introduces a novel self-support mechanism that leverages iterative pseudo-labeling to enhance segmentation accuracy in few-shot scenarios.
• It employs a self-training approach to refine segmentation masks without relying solely on scarce annotated data.
• Experimental results on benchmark datasets demonstrate significant performance improvements over previous few-shot segmentation methods.

Overview of ECCV Submission Guidelines

The document presents comprehensive guidelines for authors submitting papers to the European Conference on Computer Vision (ECCV). While it primarily focuses on formatting specifics for ECCV 2022, the principles elucidated are relevant for authors preparing submissions to similar scholarly events. With meticulous attention to detail, this document aims to standardize submissions to facilitate efficient review processes, maintaining consistency across the conference proceedings.

Content Requirements

The document is structured into clear sections delineating mandatory content requirements. Manuscripts must adhere to a maximum length of 14 pages, excluding references—a policy implemented to encourage concise and impactful submissions and circumvent the exclusion of pertinent references due to space constraints. This restriction has implications for authors in preparing their work to ensure adherence without compromising the contribution's integrity.

Compliance with Formatting Standards

The guidelines emphasize strict compliance with specific formatting standards, which include justified text within defined printing areas and designated font styles and sizes. Adherence to these standards enables a homogeneous appearance in the published proceedings. The prescribed use of LaTeX and the provided Springer class files further supports this consistency, though it may require authors unfamiliar with these tools to invest additional time in preparation.

Anonymity and Review Process

The document outlines the double-blind review policy, wherein both authors and reviewers remain anonymous to each other to avoid bias. Authors must follow detailed anonymization strategies, such as using third-person references to prior work and avoiding reference to institution-specific data in identifiable ways. This anonymity is pivotal in maintaining scientific objectivity across submissions. It highlights the importance of strategic reference management where authors are permitted to cite their previous work without revealing their identities.

Dual Submissions and Publishing Integrity

Authors are obliged to confirm that their submissions to ECCV are not concurrently under review or previously published elsewhere. This policy upholds the integrity of academic publishing and ensures that ECCV remains a premier venue for novel research in computer vision. The outlined definition of "publication" underscores the nature of peer-reviewed content versus preprints or technical reports, providing clarity on acceptable submissions.

Technical and Practical Considerations

The paper provides thorough instructions on preparing digital content, with particular attention given to the resolution and readability of figures and the consistent formatting of references. Authors are expected to submit camera-ready versions in specific file formats, alongside any required supplementary material, complying with precise submission guidelines designed to facilitate efficient typesetting and eventual distribution.

Conclusion and Implications

This guideline document, while detailed and exhaustive, is inherently procedural. Its implementation advances the quality and clarity of submitted content, fostering a coherent compilation of scholarly contributions that are accessible and engaging to the computer vision community. Looking forward, these guidelines will likely evolve with technological advances, particularly in facilitating more intuitive submission processes and perhaps integrating more flexible guidelines that accommodate various types of research contributions. Future developments may involve advanced systems for detecting formatting errors early in the submission process or ensuring consistency in anonymity in increasingly collaborative research environments.