Few-Shot Segmentation Without Meta-Learning: A Good Transductive Inference Is All You Need? (2012.06166v2)

Abstract: We show that the way inference is performed in few-shot segmentation tasks has a substantial effect on performances -- an aspect often overlooked in the literature in favor of the meta-learning paradigm. We introduce a transductive inference for a given query image, leveraging the statistics of its unlabeled pixels, by optimizing a new loss containing three complementary terms: i) the cross-entropy on the labeled support pixels; ii) the Shannon entropy of the posteriors on the unlabeled query-image pixels; and iii) a global KL-divergence regularizer based on the proportion of the predicted foreground. As our inference uses a simple linear classifier of the extracted features, its computational load is comparable to inductive inference and can be used on top of any base training. Foregoing episodic training and using only standard cross-entropy training on the base classes, our inference yields competitive performances on standard benchmarks in the 1-shot scenarios. As the number of available shots increases, the gap in performances widens: on PASCAL-5i, our method brings about 5% and 6% improvements over the state-of-the-art, in the 5- and 10-shot scenarios, respectively. Furthermore, we introduce a new setting that includes domain shifts, where the base and novel classes are drawn from different datasets. Our method achieves the best performances in this more realistic setting. Our code is freely available online: https://github.com/mboudiaf/RePRI-for-Few-Shot-Segmentation.

Overview of \LaTeX\ Author Guidelines for CVPR Proceedings

The paper "\LaTeX\ Author Guidelines for CVPR Proceedings" provides a detailed set of instructions for authors intending to submit manuscripts to the CVPR conference organized by the IEEE Computer Society. This document emphasizes the importance of adhering to specific formatting and style guidelines to ensure consistency and quality across submissions. The guidelines are framed to address common issues encountered by authors during manuscript preparation, facilitating easier processing and admission into the conference proceedings.

Core Structure and Formatting

1. Language and Submission Specifics: All manuscripts must be composed in English. The guidelines outline strict adherence to policies on dual submissions and specify that papers must not exceed eight pages, excluding references. Overlength papers face non-review, highlighting the necessity of concise and precise presentation of research.
2. Document Formatting: The recommended submission format is a two-column layout. The paper provides explicit margins and dimensions for printed material, specifying the dimensions for text areas, column widths, and space between columns to ensure uniformity in appearance.
3. Blind Review Process: The guidance discusses intuitive approaches to anonymizing submissions for blind review. It underlines that citations to one's previous work are permitted, but authors must avoid personal pronouns like "my" or "our" when discussing prior contributions, as this upholds anonymity through the review process.
4. Figures, Tables, and Equations: Instructions are provided for embedding and sizing graphical content efficiently, recommending the \verb+\includegraphics+ command for \LaTeX\ users. It remains critical that all such content maintains legibility in printed formats, considering reviewers might opt for physical copies to assess submissions.
5. Mathematics and References: The document stresses consistent numbering of equations and sections to facilitate readability and referencing. The paper advocates using the \textit{ieee_fullname} bibliographic style, which mandates precise citation practices in the research community.

Implications for Researchers

These guidelines play a crucial role in maintaining a standardized review process by ensuring uniform formatting and simplifying the logistical efforts involved in manuscript handling and evaluation. For researchers, adherence to these guidelines signifies an understanding and respect for community practices, ultimately reflecting scholarly discipline. Moreover, by eliminating formatting errors and inconsistencies, researchers can better communicate their scientific contributions without extraneous distractions.

Future Considerations

The articulated instructions signify evolving practices in the manuscript submission process for technical conferences. As methods of publication and review advance, further refinements of these guidelines could incorporate newer technologies or tools to streamline processes, such as automated formatting checks or more robust submission management systems. Authors should remain attuned to updates in these guidelines, reflecting broader shifts in academic communication and publication standards.

In conclusion, the \LaTeX\ Author Guidelines for CVPR Proceedings document establishes a comprehensive framework for manuscript preparation, enshrining practices that enhance the integrity and readability of submitted papers. This aids not only in the review and selection process but also promotes an equitable platform where scientific discourse can advance unimpeded by presentation discrepancies.