3D Model Retrieval Based On Triplet-CNN

3D model data have more visual information than 2D image data,3D model data is of great value.3D models are applied to many fields,such as biology,medical treatment,games,movie production,etc.High speed development of 3D modeling technology,The number of 3D models is growing,To reduce the cost of remodeling,Improving production efficiency at the same time,We can make use of 3D model data that has been built,Then we have to face a challenge:How to find the target model from 3D model library,This has become a hot issue that concerned.sketch is a universal method of human-computer interaction,Using sketch as input of retrieval system is more suitable for people's behavior,The research topic of 3D model based on sketch retrieval has emerged.The structure of the 3D model is too cumbersome and the uncertainty of an object representation by a sketch,It is very difficult to retrieve a 3D model from a sketch.The current mainstream sketch retrieval 3D model method uses visual feature coding,which leads to lack of learning and expressive ability,And the feature dimension is too high,which affects the speed of retrieval,More important is that the 3D retrieval of the sketch belongs to the cross domain matching problem.After understanding the problems of 3D model retrieval,This article will combine convolutional neural network,metric learning and hash coding all together,This paper presents an new algorithm,First,To solve the problem of identifying sketches and projection drawings,In this paper,we use convolutional neural network to improve the ability of feature learning,extracting the deep features of the sketch and the 3D model projection,Increasing the distinctiveness and expressive ability of features,Secondly,the similarity between sketch and 3D model projection can not be directly calculated,metric learning has the ability to learn similarity between samples,the inner relationship between the sketch and the projection can be excavated by metric learning,finally,the retrieval problem on large scale 3D model database,hash technology is widely used in retrieval tasks,In order to apply this article algorithm on a large dataset,this paper also introduces hash coding technology,the aim is to improve the speed of retrieval.