Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Adaptive Focus for Efficient Video Recognition (2105.03245v2)

Published 7 May 2021 in cs.CV, cs.LG, and eess.IV

Abstract: In this paper, we explore the spatial redundancy in video recognition with the aim to improve the computational efficiency. It is observed that the most informative region in each frame of a video is usually a small image patch, which shifts smoothly across frames. Therefore, we model the patch localization problem as a sequential decision task, and propose a reinforcement learning based approach for efficient spatially adaptive video recognition (AdaFocus). In specific, a light-weighted ConvNet is first adopted to quickly process the full video sequence, whose features are used by a recurrent policy network to localize the most task-relevant regions. Then the selected patches are inferred by a high-capacity network for the final prediction. During offline inference, once the informative patch sequence has been generated, the bulk of computation can be done in parallel, and is efficient on modern GPU devices. In addition, we demonstrate that the proposed method can be easily extended by further considering the temporal redundancy, e.g., dynamically skipping less valuable frames. Extensive experiments on five benchmark datasets, i.e., ActivityNet, FCVID, Mini-Kinetics, Something-Something V1&V2, demonstrate that our method is significantly more efficient than the competitive baselines. Code is available at https://github.com/blackfeather-wang/AdaFocus.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Yulin Wang (45 papers)
  2. Zhaoxi Chen (49 papers)
  3. Haojun Jiang (13 papers)
  4. Shiji Song (103 papers)
  5. Yizeng Han (33 papers)
  6. Gao Huang (178 papers)
Citations (95)

Summary

Overview of Author Guidelines for ICCV Proceedings

The paper "Author Guidelines for ICCV Proceedings" is a meticulous and comprehensive guide aimed at assisting authors in preparing their manuscripts for submission to the IEEE International Conference on Computer Vision (ICCV). The document outlines a variety of formatting and stylistic considerations necessary to meet the ICCV's submission standards. This paper does not introduce novel algorithms or datasets but serves as a vital tool for ensuring that submissions adhere to the conference's established requirements, thus facilitating the peer review process.

Key Guidelines

  • Abstract and Main Text: The abstract must be formatted in fully-justified italicized text and conform to specific typographical standards. The main body of the manuscript is required to be formatted in a two-column layout using 10-point Times font, ensuring consistent readability and presentation across submissions.
  • Paper Length and Format: Limits are strictly enforced, with submissions not exceeding eight pages excluding references. There is an emphasis on the correct use of LaTeX for formatting figures, equations, and text to comply with margin and font specifications.
  • Blind Review Process: The authors clarify the importance of anonymizing submissions for the double-blind review process. It is stressed that citations to one's own work should not disclose the author's identity, maintaining the integrity of the review process.
  • Figures and Tables: Guidelines emphasize the importance of making figures and tables legible in print. Authors are advised to ensure consistency in font size and line widths, accommodating readers who may prefer printed copies over electronic ones.
  • References: References should be formatted in a specific manner, with all cited works listed at the paper's end in 9-point Times font. This section doesn't contribute to the page limit, allowing for comprehensive inclusion of related works.

Implications of the Guidelines

These guidelines serve as an essential framework not only ensuring uniformity across submissions but also enhancing the accessibility and comprehensibility of presented research. A well-structured document aligns with these formal standards allowing reviewers and readers to focus on the content rather than be distracted by formatting inconsistencies. Authors are also encouraged to adhere to these guidelines to avoid desk rejection due to formatting issues, allowing their work to be evaluated purely on its scientific merit.

Future Considerations

While the paper primarily addresses formatting and submission concerns, it highlights ongoing challenges in academic publishing, such as the balance between manuscript length and content depth, the transparency of the review process, and the adaptation of guidelines to accommodate evolving digital publishing methodologies. As computer vision research continues to grow, future revisions to author guidelines may integrate technological advancements, such as interactive content or enriched metadata, aligning the dissemination of research with contemporary digital standards.

In conclusion, while the paper does not present empirical research or methodological innovations, its role in shaping the presentation and dissemination of computer vision research is invaluable. Adhering to these guidelines ensures that authors can effectively communicate their contributions to the computer vision community, facilitating a uniform and professional standard of scholarly communication.