SplatFace: Gaussian Splat Face Reconstruction Leveraging an Optimizable Surface
(2403.18784)Abstract
We present SplatFace, a novel Gaussian splatting framework designed for 3D human face reconstruction without reliance on accurate pre-determined geometry. Our method is designed to simultaneously deliver both high-quality novel view rendering and accurate 3D mesh reconstructions. We incorporate a generic 3D Morphable Model (3DMM) to provide a surface geometric structure, making it possible to reconstruct faces with a limited set of input images. We introduce a joint optimization strategy that refines both the Gaussians and the morphable surface through a synergistic non-rigid alignment process. A novel distance metric, splat-to-surface, is proposed to improve alignment by considering both the Gaussian position and covariance. The surface information is also utilized to incorporate a world-space densification process, resulting in superior reconstruction quality. Our experimental analysis demonstrates that the proposed method is competitive with both other Gaussian splatting techniques in novel view synthesis and other 3D reconstruction methods in producing 3D face meshes with high geometric precision.
Overview
-
The document serves as an instructional guide for authors submitting papers to the European Conference on Computer Vision (ECCV), detailing the submission format and review process.
-
It emphasizes adherence to the Springer LNCS style, anonymity preservation during the double-blind review process, and the requirement of line numbers in review versions.
-
Submission guidelines include the manuscript's length, the necessity of English and LaTeX format, and the prohibition of author identification to uphold double-blind review integrity.
-
The document also highlights the importance of ethical research practices, proper presentation of program code, and the structuring of citations and references according to LNCS standards.
Author Guidelines for ECCV Submission
Introduction
The document presents itself as an instructional guide for authors aiming to submit papers to the European Conference on Computer Vision (ECCV). It details the submission format, emphasizing adherence to the Springer LNCS (Lecture Notes in Computer Science) style, underlining the importance of the review process's distinct stages, and encapsulating ECCV's policies concerning paper submissions. The text serves as both an example and a comprehensive manual for preparing manuscripts for submission, highlighting key areas such as paper length, anonymity preservation, and the requisite inclusion of line numbers in review versions.
Submission Guidelines
Manuscript Requirements
The guidelines stipulate that manuscripts must be penned in English, with a strong encouragement for the LaTeX format utilizing the ECCV's official template. This template ensures conformity to Springer's typesetting requirements, incorporating line numbers for the review phase and anonymizing details to uphold the double-blind review protocol.
- Template and Formatting Conformity: Utilizing the official LNCS style is non-negotiable, with any deviation, especially in manipulating layout settings or font sizes, potentially leading to outright rejection without review.
- Paper Length and ID: Submissions must be comprehensive yet confined to 14 pages, excluding references. Each page must include the paper ID in a specified manner to facilitate the review process.
Double-Blind Review Preparation
A significant aspect of the submission process involves preparing the manuscript for double-blind review. Detailed instructions caution against any form of author identification, extending to the citation style for one's prior work and engaging with concurrent submissions. The document provides explicit examples of both poor and acceptable practices concerning anonymity, aiming to prevent inadvertent self-disclosure.
- Citing Own Work: Authors are advised to reference their work in the third person and include all necessary citations without revealing their identity.
- Handling Concurrent Submissions: The guidelines meticulously describe how to deal with overlapping submissions within the same conference timeline, emphasizing the use of supplemental material for anonymized concurrent submissions.
Formatting and Style
Visual and Structural Elements
The guidelines cover extensive formatting instructions regarding headings, figures, tables, and equations to guarantee a uniform and professional presentation of the manuscript.
- Headings should follow a specific font size and style hierarchy, strictly adhering to Springer's LNCS format.
- Figures and Tables must be clear, with captions placed correctly to facilitate comprehension. Special attention is given to the resolution and legibility of illustrations.
- Equations are to be centered, separately lineated, and sequentially numbered, highlighting the necessity for easy reference and review.
Ethical and Technical Considerations
Additional sections shed light on the importance of ethical considerations in research, the correct presentation of program code, and the structuring of citations and references per the LNCS standards. Emphasis is placed on the inclusion of Digital Object Identifiers (DOIs) to enhance the reliability and traceability of cited work.
Conclusion and Policies
The document concludes with an overview of ECCV's policies regarding submission, including aspects related to dual submissions, plagiarism, and ethical standards. Authors are reminded that acceptance for presentation and publication hinges not only on the novelty and significance of their research but equally on strict adherence to the submission guidelines.
Implications and Future Directions
The outlined submission guidelines play a pivotal role in maintaining the high standards of scholarly work presented at ECCV. By ensuring consistency in manuscript formatting and upholding the integrity of the double-blind review process, these guidelines help in advancing the field of computer vision. Future improvements in submission and review processes, potentially incorporating AI-assisted tools for checking compliance and anonymization, could further streamline submissions, enhancing the focus on scientific innovation and discovery.