Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Cross-domain three-dimensional model retrieval method based on multi-level feature alignment network

A three-dimensional model, feature pair technology, applied in digital data information retrieval, special data processing applications, image data processing and other directions, to achieve the effect of improving accuracy and retrieval accuracy

Pending Publication Date: 2020-04-28
TIANJIN UNIV
View PDF2 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] Although a lot of work has been done in the field of 3D model

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Cross-domain three-dimensional model retrieval method based on multi-level feature alignment network
  • Cross-domain three-dimensional model retrieval method based on multi-level feature alignment network
  • Cross-domain three-dimensional model retrieval method based on multi-level feature alignment network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0031] A cross-domain 3D model retrieval method based on multi-level feature alignment network, see figure 1 , the method includes the following steps:

[0032] 101: First, use a virtual camera to take virtual photos of the models in the 3D model database to generate multi-view data;

[0033] 102: A new network framework is designed: a multi-level feature alignment network for feature alignment of images and 3D models at the domain level and class level.

[0034] Among them, the feature extraction part of the multi-level feature alignment network adopts the classic CNN network (AlexNet, etc.) as the basic network structure. When extracting image features and 3D model multi-view features, the two parts of the CNN network share parameters, and then adopt a pooling layer to integrate the features of all views in the 3D model multi-view into a 3D descriptor as the feature representation of the 3D model.

[0035] 103: Multi-level feature alignment network. In order to better extr...

Embodiment 2

[0039] The scheme in embodiment 1 is further introduced below in conjunction with specific examples and calculation formulas, see the following description for details:

[0040] 201: First use the virtual camera to take virtual photos of the models in the 3D model database to generate multi-view data;

[0041] Wherein, the above step 201 mainly includes:

[0042] A set of viewpoints is predefined, and the viewpoints are the viewpoints for observing the target object. Let M be the number of predefined viewpoints. In the embodiment of the present invention, M is taken as 12, that is, a virtual For the camera, the viewpoints are completely evenly distributed around the target object. By picking different interval angles, different sets of views of the model can be obtained.

[0043] All the objects in the 3D model database are projected to obtain a set of views for each target, and a set of views of all targets constitutes a multi-view model database.

[0044] 202: Design a ne...

Embodiment 3

[0065] Below in conjunction with concrete test, the scheme in embodiment 1 and 2 is carried out feasibility verification, see the following description for details:

[0066] Using six commonly used evaluation indicators in 3D models [9] : NN, FT, ST, F, DCG, ANMMR to measure the performance of this method.

[0067] The implementation methods of these six evaluation criteria are as follows:

[0068] NN: Calculate the percentage of the nearest matching model that belongs to the query category;

[0069] FT: Calculate the recall rate of the first K relevant matching samples, where K is the cardinality of the query category;

[0070] ST: Calculate the recall rate of the first 2K relevant matching samples;

[0071] F: Comprehensive measurement of precision and recall for a certain number of retrieval results;

[0072] DCG: assign higher weights to 3D models related to the query image and lower weights to irrelevant 3D models, and then perform statistical measurements;

[0073] ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a cross-domain three-dimensional model retrieval method based on a multi-level feature alignment network, and the method comprises the steps: carrying out virtual photographingof a three-dimensional model in a three-dimensional model database through a virtual camera, and generating multi-view data; constructing a multi-level feature alignment network; performing domain-level feature alignment on an image and the features of the three-dimensional model by s discriminator through an alignment network; calculating centroids of all features of each class in the two domains of the image and the three-dimensional model based on the alignment network, taking the distance between the centroids of the same class as a part of a loss function, and minimizing the loss function by using a back propagation algorithm; and when the loss function is minimized, respectively extracting features of the image and the three-dimensional model by utilizing the trained multi-level feature alignment network, and carrying out cross-domain three-dimensional model retrieval. According to the method, a new network framework, namely a multi-level feature alignment network, is provided,feature alignment is carried out on an image domain and a three-dimensional model domain on the domain level and the class level, and the cross-domain three-dimensional model retrieval precision is improved.

Description

technical field [0001] The invention relates to the fields of feature alignment, cross-domain learning, and model retrieval, in particular to a cross-domain three-dimensional model retrieval method based on a multi-level feature alignment network. Background technique [0002] With the development of technology, the application of 3D models is becoming more and more extensive, virtual reality, augmented reality, 3D modeling [1] And other technologies are becoming more and more mature, and the data form of 3D models has shown explosive growth. The efficient and fast retrieval and classification of 3D models has also become a very important research problem. [2] . However, many previous studies and methods used 3D models as the query target. With the increase of user needs, the retrieval methods have become diverse, and the query target is not limited to the 3D model itself. The frequency of the query target in the form of images also greatly increased. However, simple feat...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06F16/532G06F16/583G06T7/66
CPCG06F16/532G06F16/583G06T7/66
Inventor 刘安安郭富宾周河宇宋丹聂为之
Owner TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com