ChamNet: Towards Efficient Network Design through Platform-Aware Model Adaptation

Abstract: This paper proposes an efficient neural network (NN) architecture design methodology called Chameleon that honors given resource constraints. Instead of developing new building blocks or using computationally-intensive reinforcement learning algorithms, our approach leverages existing efficient network building blocks and focuses on exploiting hardware traits and adapting computation resources to fit target latency and/or energy constraints. We formulate platform-aware NN architecture search in an optimization framework and propose a novel algorithm to search for optimal architectures aided by efficient accuracy and resource (latency and/or energy) predictors. At the core of our algorithm lies an accuracy predictor built atop Gaussian Process with Bayesian optimization for iterative sampling. With a one-time building cost for the predictors, our algorithm produces state-of-the-art model architectures on different platforms under given constraints in just minutes. Our results show that adapting computation resources to building blocks is critical to model performance. Without the addition of any bells and whistles, our models achieve significant accuracy improvements against state-of-the-art hand-crafted and automatically designed architectures. We achieve 73.8% and 75.3% top-1 accuracy on ImageNet at 20ms latency on a mobile CPU and DSP. At reduced latency, our models achieve up to 8.5% (4.8%) and 6.6% (9.3%) absolute top-1 accuracy improvements compared to MobileNetV2 and MnasNet, respectively, on a mobile CPU (DSP), and 2.7% (4.6%) and 5.6% (2.6%) accuracy gains over ResNet-101 and ResNet-152, respectively, on an Nvidia GPU (Intel CPU).

• The paper presents ChamNet, a framework that optimizes neural network architectures by incorporating platform-specific constraints.
• It utilizes a model adaptation strategy combined with neural architecture search to reduce latency and energy consumption.
• Experimental results validate ChamNet’s effectiveness across various hardware platforms, enabling practical deployment in resource-constrained settings.

Overview of \LaTeX\ Author Guidelines for CVPR Proceedings

This document serves as a comprehensive guide for authors submitting manuscripts to the Computer Vision and Pattern Recognition (CVPR) conference, specifically focusing on the use of \LaTeX\ for document preparation. This paper delineates essential aspects of formatting, submission requirements, and compliance guidelines that authors must adhere to in order to ensure their papers are reviewed and considered for inclusion in the conference proceedings.

Key Contributions

1. Formatting Specifications: The guide provides detailed instructions on formatting manuscripts using \LaTeX. It describes the appropriate font sizes, styles, and column dimensions necessary to comply with CVPR formatting requirements. Notably, the paper stipulates that all text must be in a two-column format with specific margins and spacing, ensuring uniformity across all submissions.
2. Manuscript Length and Review Process: Crucially, the paper outlines the limitations on manuscript length, emphasizing that papers must not exceed eight pages, excluding references. This constraint ensures that submissions are concise and focused. Overlength papers will be disqualified from the review process, highlighting the importance of strict adherence to these rules.
3. Blind Review Protocol: A fundamental aspect of the submission process is the blind review protocol. The document clarifies that authors should refrain from using language that would reveal their identity within the manuscript. It further provides examples of acceptable and unacceptable practices in maintaining anonymity while citing one's own work.
4. Illustrations and Graphics: Guidance on the inclusion of illustrations, graphs, and photographs is provided to ensure they are formatted correctly for both digital and printed versions. The document advises on font and line width adjustments to maintain consistency with the body text and ensure clarity of graphical elements.
5. Miscellaneous Requirements: Additional guidelines are presented for aspects such as numbering equations, handling dual submissions, and using color in submissions. These contribute to a detailed understanding of the various nuanced requirements necessary for the preparation of a compliant CVPR submission.

Implications and Future Prospects

The establishment of uniform formatting and submission standards has significant implications for the integrity and efficiency of the review process at CVPR. By providing precise guidelines, the document facilitates the reviewers' ability to evaluate submissions based on content rather than formatting discrepancies, potentially enhancing the overall quality of accepted papers.

Furthermore, these guidelines may help authors from diverse linguistic and cultural backgrounds by providing a clear framework for submission, thereby encouraging a broader range of contributions and fostering a diverse academic dialogue within the computer vision community.

Future developments may include updates to these guidelines to accommodate evolving technological advancements and submission platforms, potentially integrating more interactive or multimedia elements as part of the review process. Facilitating automated compliance checking tools could also streamline the submission process for authors while minimizing manual formatting errors.

In conclusion, this paper plays an integral role in the submission process for CVPR by offering precise, actionable guidelines that ensure consistency and fairness in manuscript preparation and evaluation. The adherence to these guidelines is pivotal in maintaining the conference's standard of excellence and fostering innovation in computer vision research.