Every Pixel Counts ++: Joint Learning of Geometry and Motion with 3D Holistic Understanding (1810.06125v2)

Abstract: Learning to estimate 3D geometry in a single frame and optical flow from consecutive frames by watching unlabeled videos via deep convolutional network has made significant progress recently. Current state-of-the-art (SoTA) methods treat the two tasks independently. One typical assumption of the existing depth estimation methods is that the scenes contain no independent moving objects. while object moving could be easily modeled using optical flow. In this paper, we propose to address the two tasks as a whole, i.e. to jointly understand per-pixel 3D geometry and motion. This eliminates the need of static scene assumption and enforces the inherent geometrical consistency during the learning process, yielding significantly improved results for both tasks. We call our method as "Every Pixel Counts++" or "EPC++". Specifically, during training, given two consecutive frames from a video, we adopt three parallel networks to predict the camera motion (MotionNet), dense depth map (DepthNet), and per-pixel optical flow between two frames (OptFlowNet) respectively. The three types of information are fed into a holistic 3D motion parser (HMP), and per-pixel 3D motion of both rigid background and moving objects are disentangled and recovered. Comprehensive experiments were conducted on datasets with different scenes, including driving scenario (KITTI 2012 and KITTI 2015 datasets), mixed outdoor/indoor scenes (Make3D) and synthetic animation (MPI Sintel dataset). Performance on the five tasks of depth estimation, optical flow estimation, odometry, moving object segmentation and scene flow estimation shows that our approach outperforms other SoTA methods. Code will be available at: https://github.com/chenxuluo/EPC.

Overview of "Bare Advanced Demo of IEEEtran.cls for IEEE Computer Society Journals"

The paper "Bare Advanced Demo of IEEEtran.cls for IEEE Computer Society Journals" provides a structured demonstration of the intricacies involved in preparing a manuscript for submission to IEEE Computer Society journals using the IEEEtran class in LaTeX. The document is a template that serves as a utility for authors preparing their papers and aims to streamline the formatting process according to IEEE standards.

Content and Structure

The paper exemplifies the IEEEtran.cls version 1.8b, detailing its application for compiling documents intended for IEEE Computer Society publications. This version introduces several features aimed at ensuring compliance with IEEE's rigorous publication standards, thereby facilitating both authors and publishers in maintaining consistency across different submissions. This consistency is crucial to uphold the professional aesthetics that IEEE journals are renowned for.

Structured across various sections, the template covers fundamental components of a typical IEEE paper format. Key sections include:

• Title and Author Information: It showcases the appropriate way to organize and present author affiliations and acknowledgments.
• Abstract and Keywords: It highlights the importance of the abstract and keywords in improving the discoverability and impact of the research through databases.
• Introduction and Conclusion: These foundational sections guide authors on how to effectively introduce and conclude their research.
• Appendices and References: It provides direction on how to include supplementary materials and format references in the IEEE style, which is critical for academic integrity and citation management.

Technical Implementation

The document meticulously provides directives on utilizing LaTeX commands and environments specific to the IEEEtran class. Containing examples and placeholders, the template aids authors by proposing correct LaTeX syntax, thereby minimizing formatting errors commonly encountered by those new to LaTeX or IEEE formatting specifications.

Practical Implications

From a practical standpoint, this paper, or rather the template, acts as an indispensable tool for researchers across engineering and computer science disciplines who intend to publish in IEEE journals. By adhering to this template, authors can focus more on the substance of their findings rather than spending time on the nuances of formatting, thereby potentially increasing the efficiency of the publication process.

Future Developments

As academic publication standards evolve, there is an anticipated need for continual updates to the IEEEtran class to accommodate new typesetting practices, multimedia integration, and perhaps evolving peer-review methodologies that could impact document preparation. Furthermore, with the increased adoption of automated submission systems, future versions may incorporate features that facilitate smoother interactions with these platforms.

In closing, while the primary aim of this document is to aid in formatting, its implicit contribution is in encouraging adherence to high publication standards, which ultimately supports the dissemination and credibility of scholarly work within the scientific community. As LaTeX and IEEE publication standards continue to evolve, tools such as this template will remain pivotal in bridging the gap between authors' research and its ultimate disclosure to the world.