实时可扩展的3D形状检索引擎：GIFT

132 浏览量更新于2024-08-29 收藏 889KB PDF 举报

GIFT: A Real-Time and Scalable 3D Shape Search Engine 是一篇重要的研究论文，针对三维形状检索领域提出了创新性的解决方案。人类对3D形状的感知主要依赖于从不同视角观察到的2D投影图像，这使得投影分析在3D形状检索中扮演着关键角色。然而，现有的基于投影的方法往往面临计算成本高昂的问题，无法满足大规模搜索引擎的实时性和扩展性需求。该论文的核心贡献在于构建了一个实时的3D形状搜索引擎，通过以下几个关键策略实现高效和可扩展性： 1. **GPU加速的投影与视图特征提取**：为了提高处理速度，论文提出了一种利用图形处理器（GPU）进行高效的投影和视图特征提取技术。GPU的优势在于并行处理能力，这显著提升了计算性能，使得系统能够在实时条件下处理大量数据。 2. **F-IF（First Inverted File）的引入**：为了加速多视图匹配过程，F-IF（第一倒排索引）被应用于搜索算法中。这种数据结构设计减少了查找时间，提高了匹配速度，使得查询响应时间大大缩短，满足了实时查询的需求。 3. **优化的存储和检索策略**：通过对数据进行有效的组织和存储，论文解决了传统方法中的效率瓶颈，确保了搜索引擎在处理大规模数据集时仍能保持良好的性能。 4. **可扩展性设计**：GIFT搜索引擎的设计充分考虑了系统的可扩展性，通过模块化和分布式架构，能够轻松地适应不断增长的数据量和查询需求，保证了系统的长期稳定运行。 5. **实验验证与评估**：论文通过详尽的实验和对比分析，展示了GIFT在实时性和检索精度上的优势，证明了其在实际应用中的可行性和有效性。 GIFT是一项具有革新性的技术，它不仅提高了3D形状检索的实时性能，还兼顾了系统的可扩展性，对于推动3D计算机视觉、人工智能和大数据处理等领域的发展具有重要意义。在未来的研究和实际应用中，GIFT有望成为高效的3D形状搜索解决方案的首选。

GIFT: A Real-time and Scalable 3D Shape Search Engine

Song Bai

Xiang Bai

Zhichao Zhou

Zhaoxiang Zhang

Longin Jan Latecki

Huazhong University of Science and Technology,

Temple University

CAS Center for Excellence in Brain Science and Intelligence Technology, CASIA

{songbai,xbai,zzc}@hust.edu.cn zhaoxiang.zhang@ia.ac.cn latecki@temple.edu

Abstract

Projective analysis is an important solution for 3D shape

retrieval, since human visual perceptions of 3D shapes rely

on various 2D observations from different view points. Al-

though multiple informative and discriminative views are

utilized, most projection-based retrieval systems suffer from

heavy computational cost, thus cannot satisfy the basic re-

quirement of scalability for search engines.

In this paper, we present a real-time 3D shape search

engine based on the projective images of 3D shapes. The

real-time property of our search engine results from the fol-

lowing aspects: (1) efﬁcient projection and view feature ex-

traction using GPU acceleration; (2) the ﬁrst inverted ﬁle,

referred as F-IF, is utilized to speed up the procedure of

multi-view matching; (3) the second inverted ﬁle (S-IF),

which captures a local distribution of 3D shapes in the

feature manifold, is adopted for efﬁcient context-based re-

ranking. As a result, for each query the retrieval task can

be ﬁnished within one second despite the necessary cost of

IO overhead. We name the proposed 3D shape search en-

gine, which combines GPU acceleration and Inverted File

(Twice), as GIFT. Besides its high efﬁciency, GIFT also

outperforms the state-of-the-art methods signiﬁcantly in re-

trieval accuracy on various shape benchmarks and compe-

titions.

1. Introduction

3D shape retrieval is a fundamental issue in computer

vision and pattern recognition. With the rapid develop-

ment of large scale public 3D repositories, e.g., Google 3D

Warehouse or TurboSquid, and large scale shape bench-

marks, e.g., ModelNet [39], SHape REtrieval Contest

(SHREC) [14, 31], the scalability of 3D shape retrieval al-

gorithms becomes increasingly important for practical ap-

plications. However, efﬁciency issue has been more or less

ignored by previous works, though enormous efforts have

been devoted to retrieval effectiveness, that is to say, to de-

sign informative and discriminative features [12, 2, 17, 6,

40, 15, 18] to boost the retrieval accuracy. As suggested

in [14], plenty of these algorithms do not scale up to large

3D shape databases due to their high time complexity.

Meanwhile, owing to the fact that human visual percep-

tion of 3D shapes depends upon 2D observations, projective

analysis has became a basic and inherent tool in 3D shape

domain for a long time, with applications to segmenta-

tion [38], matching [24], reconstruction, etc.. Speciﬁcally in

3D shape retrieval, projection-based methods demonstrate

impressive performances. Especially in recent years, the

success of planar image representation [7, 35, 43], makes it

easier to describe 3D models using depth or silhouette pro-

jections.

Generally, a typical 3D shape search engine is comprised

of the following four components (see also Fig. 1):

1. Projection rendering. With a 3D model as input, the

output of this component is a collection of projec-

tions. Most methods set an array of virtual cameras

at pre-deﬁned view points to capture views. These

view points can be the vertices of a dodecahedron [4],

located on the unit sphere [35], or around the lateral

surface of a cylinder [24]. In most cases, pose nor-

malization [22] is needed for the sake of invariance to

translation, rotation and scale changes.

2. View feature extraction. The role of this component is

to obtain multiple view representations, which affects

the retrieval quality largely. A widely-used paradigm

is Bag-of-Words (BoW) [7] model, since it has shown

its superiority as natural image descriptors. However,

in order to get better performances, many features [14]

are of extremely high dimension. As a consequence,

raw descriptor extraction (e.g., SIFT [20]), quantiza-

tion and distance calculation are all time-consuming.

3. Multi-view matching. This component establishes the

correspondence between two sets of view features, and

returns a matching cost between two 3D models. Since

at least a set-to-set matching strategy [25, 26, 27, 16, 9]

is required, this stage suffers from high time complex-

ity even when using the simplest Hausdorff matching.

Hence, the usage of algorithms incorporated with some

下载后可阅读完整内容，剩余9页未读，立即下载

weixin_38724247

粉丝: 8
资源: 915

实时可扩展的3D形状检索引擎：GIFT

Practical Node.js: Building Real-World Scalable Web Apps, 2nd Edition

Practical Node.js: Building Real-World Scalable Web Apps

Real-Time Adaptive Scalable TextureCompression for the Web.pdf

Real-Time Design Patterns - Robust Scalable Architecture for Real-Time

GIFT: Towards Scalable 3D Shape Retrieval

Big Data - PRINCIPLES AND BEST PRACTICES OF SCALABLE REAL-TIME DATA SYSTEMS

Cloud Native Architectures: Design high-availability and cost-effective

intel/Reference-PE-and-Measurements-DB-for-WiFi-Time-based-Scalable-Location:基于 WiFi 时间的可扩展定位的参考定位引擎和测量数据库-matlab开发

【Advanced Chapter】Advanced Web Crawler Practices: Crawling Dynamic Webpage Data: Real-time Data ...

A scalable storage supporting multistream real-time data retrieval

最新资源