Triplet-Center Loss for Multi-view 3D Object Retrieval

  title={Triplet-Center Loss for Multi-view 3D Object Retrieval},
  author={Xinwei He and Yang Zhou and Zhichao Zhou and Song Bai and Xiang Bai},
  journal={2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  • Xinwei He, Yang Zhou, X. Bai
  • Published 16 March 2018
  • Computer Science
  • 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition
Most existing 3D object recognition algorithms focus on leveraging the strong discriminative power of deep learning models with softmax loss for the classification of 3D data, while learning discriminative features with deep metric learning for 3D object retrieval is more or less neglected. In the paper, we study variants of deep metric learning losses for 3D object retrieval, which did not receive enough attention from this area. First, two kinds of representative losses, triplet loss and… 

Figures and Tables from this paper

Center-push loss for joint view-based 3D model classification and retrieval feature learning

It is argued that although the classification and retrieval focus on different characteristics of embedding features, they are compatible rather than opposed to each other, and a novelty loss named center-push loss for joint feature learning is proposed.

Angular Triplet-Center Loss for Multi-view 3D Shape Retrieval

A novel metric loss named angular triplet-center loss is proposed, which directly optimizes the cosine distances between the features of 3D shape features to achieve larger inter-class distance and smaller intra- class distance simultaneously.

An Improved Multi-View Convolutional Neural Network for 3D Object Retrieval

An improved Multi-view Convolutional Neural Network (MVCNN) for view-based 3D object representation learning is presented and Group-view Similarity Learning (GSL) is proposed over the multi-view representations before the aggregation operation (i.e., max-pooling in MVCNN).

Rethinking Loss Design for Large-scale 3D Shape Retrieval

The Collaborative Inner Product Loss (CIP Loss) is proposed to obtain ideal shape embedding that discriminative among different categories and clustered within the same class to be clustered in a linear subspace.

Joint Heterogeneous Feature Learning and Distribution Alignment for 2D Image-Based 3D Object Retrieval

This paper proposes to learn a mapping function in the Grassmann manifold to reduce the divergence of heterogeneous features of 2D images and 3D objects and employs the data distribution alignment method to adaptively integrate both marginal and conditional distributions.

MVPN: Multi-View Prototype Network for 3D Shape Recognition

A discriminative loss is proposed, which encourages intra-class compactness and inter-class separability between learned representations, making the representation more discriminatives and robust.

Learning generalizable deep feature using triplet-batch-center loss for person re-identification

A novel metric-learning loss function, triplet-batch-center loss (TBCL), to learn more generalizable deep features than the previous loss functions, which leads to state-of-the-art results on these datasets without the image re-ranking post-processing.

Multi-graph Convolutional Network for Unsupervised 3D Shape Retrieval

A novel multi-graph network (MGN) is proposed for unsupervised 3D shape retrieval, which utilizes the correlations among modalities and structural similarity between two models to guide the shape representation learning process without category information.

Reference-oriented Loss for Person Re-identification

A novel loss function called reference-oriented triplet loss is proposed, which introduces several reference images to guide training and distances between the reference image and images of the same identity are required to be as similar as possible.

DeepCCFV: Camera Constraint-Free Multi-View Convolutional Neural Network for 3D Object Retrieval

By reducing the over-fitting issue, a camera constraint-free multi-view convolutional neural network named DeepCCFV is constructed and the effectiveness of the proposed method in free camera settings comparing with existing state-of-theart 3D object retrieval methods is demonstrated.



DeepShape: Deep-Learned Shape Descriptor for 3D Shape Retrieval

This paper proposes a novel 3D shape feature learning method to extract high-level shape features that are insensitive to geometric deformations of shapes using a discriminative deep auto-encoder to learn deformation-invariant shape features.

Learning Barycentric Representations of 3D Shapes for Sketch-Based 3D Shape Retrieval

This paper proposes to learn barycenters of 2D projections of 3D shapes for sketch-based 3D shape retrieval by using two deep convolutional neural networks (CNNs) to extract deep features of sketches and 2D projected shapes to form a barycentric representation.

Deep Metric Learning with Angular Loss

This paper proposes a novel angular loss, which takes angle relationship into account, for learning better similarity metric, and aims at constraining the angle at the negative point of triplet triangles.

Deep Aggregation of Local 3D Geometric Features for 3D Model Retrieval

A novel deep neural network called Deep Local feature Aggregation Network (DLAN) is proposed that combines extraction of rotation-invariant 3D local features and their aggregation in a single deep architecture and Experimental evaluation shows that the DLAN outperforms the existing deep learning-based 3DMR algorithms.

A Discriminative Feature Learning Approach for Deep Face Recognition

This paper proposes a new supervision signal, called center loss, for face recognition task, which simultaneously learns a center for deep features of each class and penalizes the distances between the deep features and their corresponding class centers.

Ranking on Cross-Domain Manifold for Sketch-Based 3D Model Retrieval

Experimental evaluation by using sketch-based 3D model retrieval benchmarks showed that the proposed Cross-Domain Manifold Ranking (CDMR), an algorithm that effectively compares two sets of features that lie in different domains, is more accurate than state-of-the-art sketch- based 3D models retrieval algorithms.

Learning Cross-Domain Neural Networks for Sketch-Based 3D Shape Retrieval

Experimental results suggest that both CDNN and PCDNN can outperform state-of-the-art performance, where PCdNN can further improve CDNN when employing a hierarchical structure.

GIFT: Towards Scalable 3D Shape Retrieval

A real-time 3D shape search engine, which combines GPU acceleration and inverted file (t wice) as GIFT, which outperforms state-of-the-art methods significantly in retrieval accuracy on various shape benchmarks and competitions.

Sketch-based 3D shape retrieval using Convolutional Neural Networks

  • Fang WangLe KangYi Li
  • Computer Science
    2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2015
This work drastically reduces the number of views to only two predefined directions for the whole dataset, and learns two Siamese Convolutional Neural Networks, one for the views and one forThe sketches, which is significantly better than state of the art approaches, and outperforms them in all conventional metrics.

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

The proposed 3D shape search engine, which combines GPU acceleration and Inverted File (Twice), is named as GIFT, which outperforms the state-of-the-art methods significantly in retrieval accuracy on various shape benchmarks and competitions.