Three-dimensional model classification and retrieval method based on panorama and multi-channel CNN

A technology of three-dimensional models and panoramas, applied in biological neural network models, special data processing applications, instruments, etc., can solve problems affecting classification and search accuracy, and achieve the effect of improving robustness

Active Publication Date: 2018-12-21
TIANJIN UNIV
View PDF5 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In general, all these methods usually focus on structural information or visual information while ignoring the other, affecting classification and search accuracy

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
  • Three-dimensional model classification and retrieval method based on panorama and multi-channel CNN
  • Three-dimensional model classification and retrieval method based on panorama and multi-channel CNN
  • Three-dimensional model classification and retrieval method based on panorama and multi-channel CNN

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0032] To solve the above problems, multi-resolution panoramic views need to be extracted from each 3D model. Research shows that the panoramic view of the 3D model can convert the structural information of the 3D model into 2D image information [13] . The embodiment of the present invention proposes a 3D model view extraction method based on panorama and multi-channel CNN, see figure 1 and figure 2 , see the description below:

[0033] 101: By putting figure 2 The surface of the 3D model in is projected onto the side of a cylinder with radius R and height H=2R, to figure 2 The origin of the 3D model in the center is the center, and the axis of the 3D model is parallel to one of the main axes of X, Y, and Z in space to obtain the initial panorama;

[0034] Among them, the value of R is set to 3*d max , d max is the maximum distance from the surface of the 3D model to the centroid.

[0035] 102: Assuming that the Z-axis panorama is extracted, the embodiment of the pr...

Embodiment 2

[0042] The following combines the specific network structure, calculation formula, image 3 , Figure 4 The training of multi-channel and multi-scale CNN networks and the similarity measurement between two different 3D models are introduced, as described below:

[0043] image 3 A multi-scale network is shown. This network includes three scales to extract view descriptors of different resolutions of the same input picture. Assume that the input picture size is 256*256. For the first scale, the input picture size is the same as the original picture size , the size is 256*256, the feature map is obtained directly through the convolutional neural network of VGG16, and then the 4096-dimensional feature map is obtained after normalization processing;

[0044] For the second scale, the scale of the input image is converted to 128*128, so that the resolution of the generated image is 1 / 2 of the resolution of the original image, and then downsampling is performed, and the feature ma...

Embodiment 3

[0054] Below in conjunction with concrete data set, table 1, table 2, carry out result verification to the scheme in embodiment 1 and 2, see the following description for details:

[0055] The dataset used to evaluate the proposed classification method is the Princeton ModelNet large-scale 3DCAD model dataset. ModelNet consists of 127,915 CAD models grouped into 662 object categories into two subsets, ModelNet-10 and ModelNet-40, both of which contain training and testing partitions.

[0056] 1) ModelNet10 consists of 4899 CAD models divided into 10 categories. For ease of processing, these models are adjusted to place the center of mass of the model at the origin of the coordinates and normalized in terms of translation and rotation.

[0057] Among them, the training and testing subsets of ModelNet10 consist of 3991 and 908 models, respectively.

[0058] 2) ModelNet-40 contains 12,311 CAD models divided into 40 categories. For ease of handling, these models were adjusted t...

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 method for extracting three-dimensional model view based on panorama and multi-channel CNN. The method comprises the following steps: projecting a 3D model onto a side surface of a cylinder satisfying preset conditions; taking an origin of the 3D model as a center, obtaining an initial panorama by paralleling an axis of the 3D model to one of the principal axes of X, Y and Z; obtaining an initial panorama by projecting a 3D model onto a side surface of a cylinder satisfying preset conditions; 3-D model surface angle phi and y coordinates are sample at a preset rate, acquiring two sets of values of each point in the initial panorama to represent the position characteristics of the 3D model surface and the directional characteristics of the 3D model surface in three-dimensional space; A multi-scale network and a multi-channel convolution neural network are constructed. The 3D model surface location features and 3D model surface orientation features are used as inputs to train the network and measure the similarity between the two different 3D models. The invention retains the local and global information of the structure and vision of the three-dimensional model, automatically calculates the features of the 2D panoramic view, and is used for processing classification and retrieval problems.

Description

technical field [0001] The invention relates to the field of three-dimensional model classification and retrieval, in particular to a three-dimensional model classification and retrieval method based on panorama and CNN multi-channel CNN. Background technique [0002] With the development of computer vision technology, 3D technology is widely used in film and television industry, mechanical design, construction industry, infrastructure, entertainment industry, medical treatment and other fields. More and more people begin to upload their own 3D models on certain websites, and the number of 3D models is increasing. This has led to 3D model retrieval being a hot topic in computer vision. Different from the traditional visual representation of 2D image information, 3D models have not only visual information but also structural information. Therefore, traditional computer vision techniques are difficult to use to represent 3D models. In recent years, many methods have been pr...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06F17/30G06N3/04
CPCG06N3/045
Inventor 梁祺聂为之
Owner TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products