An Analysis of Visual Question Answering Algorithms (1703.09684v2)

Abstract: In visual question answering (VQA), an algorithm must answer text-based questions about images. While multiple datasets for VQA have been created since late 2014, they all have flaws in both their content and the way algorithms are evaluated on them. As a result, evaluation scores are inflated and predominantly determined by answering easier questions, making it difficult to compare different methods. In this paper, we analyze existing VQA algorithms using a new dataset. It contains over 1.6 million questions organized into 12 different categories. We also introduce questions that are meaningless for a given image to force a VQA system to reason about image content. We propose new evaluation schemes that compensate for over-represented question-types and make it easier to study the strengths and weaknesses of algorithms. We analyze the performance of both baseline and state-of-the-art VQA models, including multi-modal compact bilinear pooling (MCB), neural module networks, and recurrent answering units. Our experiments establish how attention helps certain categories more than others, determine which models work better than others, and explain how simple models (e.g. MLP) can surpass more complex models (MCB) by simply learning to answer large, easy question categories.

Overview of "LaTeX Author Guidelines for ICCV Proceedings"

The paper "LaTeX Author Guidelines for ICCV Proceedings" serves as a comprehensive style guide designed to streamline the submission process of manuscripts to the International Conference on Computer Vision (ICCV). Targeted at authors utilizing the \LaTeX\ document preparation system, this guide encapsulates the essential requirements that need to be adhered to for a successful manuscript submission.

The paper delineates its guidance through sections focusing on manuscript language, submission policies, paper length, and formatting details, emphasizing the importance of strict compliance with formatting requirements. These requirements include specifics on the text area dimensions, column formatting, and proper use of type styles and fonts.

Submission and Formatting Guidelines

The guidelines highlight several notable modifications from previous iterations, including the updated policy on paper length and the absence of allowances for revisions. Manuscripts must be concise, with a maximum length of eight pages excluding references. A critical compliance area involves the anti-alteration rules, where any deviation from the prescribed margins and formatting parameters restricts the paper from being reviewed.

The paper also addresses the use of a ruler in draft submissions, assisting reviewers in pinpointing sections within the text, although this is to be omitted from the final submission. The pursuit of a consistent style extends to mathematical equations which must be numbered irrespective of whether they are referenced in the text, facilitating ease of reference for future readers.

Blind Review Process

The approach to blind peer review is elaborated upon, clarifying common misconceptions. The guidelines emphasize proper citation of previous works without self-identification, thereby maintaining the anonymity integral to the blind review process. This includes avoiding personal descriptors like "our" when referring to citations of the authors' prior works.

Technical Specifications

For the technical aspects, authors are instructed on optimizing their submissions through \LaTeX-specific commands for illustrations and graphics. The paper elucidates using commands like \includegraphics for figures, ensuring they are of an appropriate resolution for both on-screen viewing and printed formats.

The use of fonts and the structure of headings are carefully articulated, specifying Times New Roman or its equivalent in various point sizes and weights according to the section hierarchy. Formalities such as the presentation of tables, captions, and their integration in the text are also addressed succinctly, enforcing uniformity across submissions.

Miscellaneous Details and Final Submission

Attention is given to ancillary details such as proper citation order and the preference for internal in-line citations for enhanced clarity. Authors are forewarned about the strategic use of footnotes, advocating in-text explanations as the primary method for additional information provision.

The necessity of submitting a signed IEEE copyright release form during the final paper submission is underscored as an obligatory step prior to publication. This is obligatory for ensuring the paper's inclusion in the conference proceedings.

Conclusion

This paper provides vital information for prospective contributors to ICCV, ensuring their submissions meet the expected standards of clarity and uniformity. While these guidelines may appear intricate, their diligent implementation will facilitate a smoother review process and enhance the quality and consistency of the conference's published content. For researchers in computer vision, adhering to these guidelines is a practical endeavor that aligns with their broader objective of contributing to the field's body of knowledge. As such, a thorough comprehension and application of these guidelines are crucial for authors aiming to have their work recognized in this competitive and influential forum. Future iterations may potentially incorporate more automation in formatting or consider evolving digital dissemination practices, but the core emphasis on rigorous formatting will likely remain unchanged.