Learning Deep Embeddings with Histogram Loss

Abstract

We suggest a loss for learning deep embeddings. The new loss does not introduce parameters that need to be tuned and results in very good embeddings across a range of datasets and problems. The loss is computed by estimating two distribution of similarities for positive (matching) and negative (non-matching) sample pairs, and then computing the probability of a positive pair to have a lower similarity score than a negative pair based on the estimated similarity distributions. We show that such operations can be performed in a simple and piecewise-differentiable manner using 1D histograms with soft assignment operations. This makes the proposed loss suitable for learning deep embeddings using stochastic optimization. In the experiments, the new loss performs favourably compared to recently proposed alternatives.

NIPS 2016

References in corpus (2)

• Deep Learning Face Representation by Joint Identification-Verification
• Deep Metric Learning for Practical Person Re-Identification

Cited by in corpus (48)

• R2D2: Repeatable and Reliable Detector and Descriptor
• Deep Adaptive Feature Embedding with Local Sample Distributions for Person Re-identification
• Circle Loss: A Unified Perspective of Pair Similarity Optimization
• Sampling Matters in Deep Embedding Learning
• Revisiting Training Strategies and Generalization Performance in Deep Metric Learning
• Domain Generalization via Optimal Transport with Metric Similarity Learning
• Optimizing Rank-based Metrics with Blackbox Differentiation
• CoMIR: Contrastive Multimodal Image Representation for Registration
• AC-VRNN: Attentive Conditional-VRNN for Multi-Future Trajectory Prediction
• Learning Pair Potentials using Differentiable Simulations
• SphereFace2: Binary Classification is All You Need for Deep Face Recognition
• Sharing Matters for Generalization in Deep Metric Learning
• MIC: Mining Interclass Characteristics for Improved Metric Learning
• Deep Metric Learning with Spherical Embedding
• Class Rectification Hard Mining for Imbalanced Deep Learning
• Discriminative structural graph classification
• Unsupervised Deep Metric Learning via Auxiliary Rotation Loss
• Learning Spread-out Local Feature Descriptors
• Divide and Conquer the Embedding Space for Metric Learning
• Deep Metric Learning Beyond Binary Supervision
• Stochastic Prototype Embeddings
• In Defense of the Triplet Loss Again: Learning Robust Person Re-Identification with Fast Approximated Triplet Loss and Label Distillation
• Multiple-Identity Image Attacks Against Face-based Identity Verification
• Robust and Decomposable Average Precision for Image Retrieval
• Deep Metric Learning Meets Deep Clustering: An Novel Unsupervised Approach for Feature Embedding
• A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning
• Person Re-identification with Metric Learning using Privileged Information
• Global Distance-distributions Separation for Unsupervised Person Re-identification
• Signal-to-Noise Ratio: A Robust Distance Metric for Deep Metric Learning
• Instance Cross Entropy for Deep Metric Learning
• Ladder Loss for Coherent Visual-Semantic Embedding
• GB-CosFace: Rethinking Softmax-based Face Recognition from the Perspective of Open Set Classification
• Improving Face Recognition from Hard Samples via Distribution Distillation Loss
• Multi-Modal Retrieval using Graph Neural Networks
• The Earth Mover's Pinball Loss: Quantiles for Histogram-Valued Regression
• Learning Portrait Style Representations
• Multi-views Embedding for Cattle Re-identification
• Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning
• DIABLO: Dictionary-based Attention Block for Deep Metric Learning
• Continuous Histogram Loss: Beyond Neural Similarity
• Atlas Based Representation and Metric Learning on Manifolds
• Adaptive Hierarchical Similarity Metric Learning with Noisy Labels
• Multi-Observation Elicitation
• Speaker diarization with session-level speaker embedding refinement using graph neural networks
• Unifying Few- and Zero-Shot Egocentric Action Recognition
• Diversified Mutual Learning for Deep Metric Learning
• Metric Learning for Dynamic Text Classification
• Robust Angular Local Descriptor Learning