Unsupervised Pre-Training of Image Features on Non-Curated Data (1905.01278v3)

Abstract: Pre-training general-purpose visual features with convolutional neural networks without relying on annotations is a challenging and important task. Most recent efforts in unsupervised feature learning have focused on either small or highly curated datasets like ImageNet, whereas using uncurated raw datasets was found to decrease the feature quality when evaluated on a transfer task. Our goal is to bridge the performance gap between unsupervised methods trained on curated data, which are costly to obtain, and massive raw datasets that are easily available. To that effect, we propose a new unsupervised approach which leverages self-supervision and clustering to capture complementary statistics from large-scale data. We validate our approach on 96 million images from YFCC100M, achieving state-of-the-art results among unsupervised methods on standard benchmarks, which confirms the potential of unsupervised learning when only uncurated data are available. We also show that pre-training a supervised VGG-16 with our method achieves 74.9% top-1 classification accuracy on the validation set of ImageNet, which is an improvement of +0.8% over the same network trained from scratch. Our code is available at https://github.com/facebookresearch/DeeperCluster.

• The paper introduces a novel unsupervised pre-training method that exploits non-curated data to learn robust image features.
• It details a scalable methodology comparing pre-training on non-curated datasets against curated ones, revealing significant performance gains.
• The study demonstrates enhanced accuracy in downstream image recognition tasks, underscoring practical viability in real-world applications.

Overview of Guidelines for ICCV Proceedings

The document titled "LaTeX Author Guidelines for ICCV Proceedings" serves as a prescriptive manual for authors preparing manuscripts for submission to the International Conference on Computer Vision (ICCV). It is of vital importance that researchers adhere to the sequential directives provided to ensure their work aligns with the imposed standards and formats mandated by the IEEE Computer Society Press.

Structural Composition and Content Requirements

The document elaborates on distinct themes that authors should be especially meticulous about: language, submission policies, document length, use of formatting tools, and anonymity in submission processes. These instructions are integral to standardizing the presentation of submissions, thereby facilitating an efficient and organized review process.

• Language and Submission Protocols: Manuscripts are required to be composed in English, and use of the LaTeX document preparation system is emphasized. It also underscores the importance of complying with dual submission policies to avoid conflicts with concurrent submission to other entities.
• Length and Formatting Constraints: The paper should not exceed eight pages, excluding references, which remain unlimited in length. Overlength submissions, altered formats, or inappropriate modifications resulting in deviation from the specified structure may forfeit the manuscript's eligibility for review.
• Review Process and Anonymity: The guidelines stress the importance of blind review practices which require the exclusion of specific author-identifying phrases, particularly when citing personal prior work. Nonetheless, the document provides strategies for correctly referencing self-authored works or concurrently submitted papers elsewhere.

Technical Specifications

The guide prescribes precise formatting rules to enhance manuscript readability and consistency.

• Use of Rulers and Mathematics: LaTeX style settings include rulers for annotative ease during reviews, contributing to clarity when specific lines or equations need referencing. Additionally, all mathematical formulations should be displayed with appropriate numbering for streamlined citation.
• Figures, Tables, and Captions: Clarity and coherence in visual representations are critical. Figures and tables should be integrated with captions using reduced font sizes and centered alignment, maintaining professional aesthetics.
• Text and Headings Format: Instructions for text and section heading styles call for a combination of Times and Times Roman fonts with specific size and boldness guidelines. First-order headings, for instance, should be specified with Times 12-point boldface.

Practical Implications and Future Considerations

Adherence to these guidelines ensures that manuscripts are presented within a universal format that simplifies the review process while upholding the standardization necessary for academic dissemination. For potential authors, understanding and applying these formatting rules is crucial not only to avoid desk rejection but also to facilitate clear communication of scientific insights.

As academia continues to evolve, formats and protocols referenced in this document may adapt with technologies and publication behaviors. Emerging authors must remain vigilant, understanding that such guidelines are subject to change, reflecting the shifting dynamics within scientific publishing norms.

Overall, this guide provides a detailed roadmap for submitting manuscripts to the ICCV, embodying the precise combination of content structure and technical formatting essential for academic writing. Authors would benefit from diligently adhering to these specifications, not only to meet publication criteria but also to enhance the communication and impact of their work.