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

D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (2003.03164v1)

Published 6 Mar 2020 in cs.CV

Abstract: A successful point cloud registration often lies on robust establishment of sparse matches through discriminative 3D local features. Despite the fast evolution of learning-based 3D feature descriptors, little attention has been drawn to the learning of 3D feature detectors, even less for a joint learning of the two tasks. In this paper, we leverage a 3D fully convolutional network for 3D point clouds, and propose a novel and practical learning mechanism that densely predicts both a detection score and a description feature for each 3D point. In particular, we propose a keypoint selection strategy that overcomes the inherent density variations of 3D point clouds, and further propose a self-supervised detector loss guided by the on-the-fly feature matching results during training. Finally, our method achieves state-of-the-art results in both indoor and outdoor scenarios, evaluated on 3DMatch and KITTI datasets, and shows its strong generalization ability on the ETH dataset. Towards practical use, we show that by adopting a reliable feature detector, sampling a smaller number of features is sufficient to achieve accurate and fast point cloud alignment.code release

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Xuyang Bai (12 papers)
  2. Zixin Luo (20 papers)
  3. Lei Zhou (126 papers)
  4. Hongbo Fu (67 papers)
  5. Long Quan (35 papers)
  6. Chiew-Lan Tai (12 papers)
Citations (335)

Summary

  • The paper introduces a unified framework that jointly learns dense 3D keypoint detection and description to enhance feature matching performance.
  • It leverages deep learning to overcome traditional limitations, demonstrating superior accuracy on benchmark datasets.
  • The method shows promising applications in robust 3D reconstruction and scene understanding in complex real-world environments.

Author Guidelines for CVPR Proceedings

The paper, titled "LaTeX Author Guidelines for CVPR Proceedings," serves as a comprehensive guide for authors submitting their research to the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR). It systematically lays out various directives essential for preparing manuscripts, presenting a clear and precise framework for authors to follow.

Structure and Formatting

The document explores the specific requirements for formatting submissions using LaTeX, emphasizing the adoption of a two-column format for all text. The allowable dimensions for the text and the detailed configuration of margins are meticulously described, indicating a text width of 6 7/8 inches and a height of 8 7/8 inches. The guide also specifies the typeface requirements, suggesting Times or Times Roman, and underscores maintaining consistency in font size throughout different sections of the paper.

Notably, the directive highlights the use of a printed ruler. This tool aids reviewers in referencing specific lines in the document, enhancing clarity in communications during the review process. The removal of this ruler on the final camera-ready copy is emphasized.

Submissions and Review Process

Attention is given to the submission guidelines, particularly regarding dual submissions and paper length limitations. Manuscripts are restricted to a maximum of eight pages, excluding references, beyond which they are not subject to review. The role of blind review is articulated, advising authors on how to appropriately anonymize references to their previous work while maintaining transparency regarding the foundation of their current research.

Technical Content

The rules for mathematical content are also addressed. Authors are instructed to number all sections and equations distinctly, facilitating easy reference in future discussions or citations. Furthermore, the inclusion of figures and tables requires a rigorous approach regarding their alignment, captioning, and font sizes, ensuring that they are legible in printed formats.

Miscellaneous Guidance

The document touches on various auxiliary elements, such as the proper use of footnotes and references. Footnotes are to be used sparingly, while references follow a specific citation style, using square brackets for in-text citations. The document also covers the usage of color and the importance of adapting graphical elements to be effective both in digital and printed formats.

Practical Implications

The implications of these guidelines are primarily operational, aiding in the standardization of submissions to streamline the review process and publication. Adhering to these guidelines ensures that all submitted papers meet an established level of quality and consistency, facilitating fair and efficient peer review.

In conclusion, this paper serves as an essential resource for authors intending to submit to CVPR, providing detailed instructions that enhance the clarity, consistency, and professionalism of their submissions. Understanding and adhering to these guidelines is crucial for authors to ensure successful participation in the CVPR submission and review process. Future iterations may further refine these guidelines as the conference evolves and as digital and publication standards advance.

Youtube Logo Streamline Icon: https://streamlinehq.com