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 (4)

• Deep Learning Face Representation by Joint Identification-Verification
• CNN Features off-the-shelf: an Astounding Baseline for Recognition
• Deep Metric Learning for Practical Person Re-Identification
• DeepHash: Getting Regularization, Depth and Fine-Tuning Right

Cited by in corpus (85)

• R2D2: Repeatable and Reliable Detector and Descriptor
• Deep Adaptive Feature Embedding with Local Sample Distributions for Person Re-identification
• Multi-Similarity Loss with General Pair Weighting for Deep Metric Learning
• 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
• Learning View-Specific Deep Networks for Person Re-Identification
• Discriminative Feature Learning with Foreground Attention for Person Re-Identification
• Domain Generalization via Optimal Transport with Metric Similarity Learning
• Optimizing Rank-based Metrics with Blackbox Differentiation
• Metric Learning for Novelty and Anomaly Detection
• CoMIR: Contrastive Multimodal Image Representation for Registration
• Learning Beyond Human Expertise with Generative Models for Dental Restorations
• Smart Mining for Deep Metric Learning
• AC-VRNN: Attentive Conditional-VRNN for Multi-Future Trajectory Prediction
• Heated-Up Softmax Embedding
• Classification is a Strong Baseline for Deep Metric Learning
• Learning Pair Potentials using Differentiable Simulations
• Unsupervised Cross-dataset Person Re-identification by Transfer Learning of Spatial-Temporal Patterns
• SphereFace2: Binary Classification is All You Need for Deep Face Recognition
• Batch DropBlock Network for Person Re-identification and Beyond
• Attention-based Ensemble for Deep Metric Learning
• Sharing Matters for Generalization in Deep Metric Learning
• From handcrafted to deep local features
• MIC: Mining Interclass Characteristics for Improved Metric Learning
• Hardness-Aware Deep Metric Learning
• Deep Metric Learning with Spherical Embedding
• Unsupervised Deep Metric Learning via Auxiliary Rotation Loss
• Class Rectification Hard Mining for Imbalanced Deep Learning
• Discriminative structural graph classification
• Hard-Aware Deeply Cascaded Embedding
• Learning Spread-out Local Feature Descriptors
• Divide and Conquer the Embedding Space for Metric Learning
• Deep Metric Learning Beyond Binary Supervision
• Stochastic Prototype Embeddings
• Deep Reinforcement Learning Attention Selection for Person Re-Identification
• Matchable Image Retrieval by Learning from Surface Reconstruction
• Multiple-Identity Image Attacks Against Face-based Identity Verification
• In Defense of the Triplet Loss Again: Learning Robust Person Re-Identification with Fast Approximated Triplet Loss and Label Distillation
• A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning
• Deep Metric Learning Meets Deep Clustering: An Novel Unsupervised Approach for Feature Embedding
• Robust and Decomposable Average Precision for Image Retrieval
• Person Re-identification with Metric Learning using Privileged Information
• An Evaluation of Deep CNN Baselines for Scene-Independent Person Re-Identification
• Global Distance-distributions Separation for Unsupervised Person Re-identification
• MIHash: Online Hashing with Mutual Information
• Signal-to-Noise Ratio: A Robust Distance Metric for Deep Metric Learning
• Deep Metric Learning with Hierarchical Triplet Loss
• Ranked List Loss for Deep Metric Learning
• Embarrassingly Simple Binary Representation Learning
• Instance Cross Entropy for Deep Metric Learning
• Deep Metric Learning by Online Soft Mining and Class-Aware Attention
• Imbalanced Deep Learning by Minority Class Incremental Rectification
• Searching for Apparel Products from Images in the Wild
• Predicting Conversion of Mild Cognitive Impairments to Alzheimer's Disease and Exploring Impact of Neuroimaging
• Improving Face Recognition from Hard Samples via Distribution Distillation Loss
• Automatic Dataset Annotation to Learn CNN Pore Description for Fingerprint Recognition
• GB-CosFace: Rethinking Softmax-based Face Recognition from the Perspective of Open Set Classification
• Ladder Loss for Coherent Visual-Semantic Embedding
• The Earth Mover's Pinball Loss: Quantiles for Histogram-Valued Regression
• Multi-Modal Retrieval using Graph Neural Networks
• Learning Joint Acoustic-Phonetic Word Embeddings
• Multi-views Embedding for Cattle Re-identification
• DIABLO: Dictionary-based Attention Block for Deep Metric Learning
• Continuous Histogram Loss: Beyond Neural Similarity
• Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning
• Scalable and Effective Deep CCA via Soft Decorrelation
• Deep Metric Learning with Density Adaptivity
• Hashing as Tie-Aware Learning to Rank
• Learning Portrait Style Representations
• Deep Randomized Ensembles for Metric Learning
• Open-Ended Content-Style Recombination Via Leakage Filtering
• Unsupervised Person Re-identification by Deep Learning Tracklet Association
• Atlas Based Representation and Metric Learning on Manifolds
• Metric Learning for Dynamic Text Classification
• Deep Embedding using Bayesian Risk Minimization with Application to Sketch Recognition
• Diversified Mutual Learning for Deep Metric Learning
• Robust Angular Local Descriptor Learning
• High Diversity Attribute Guided Face Generation with GANs
• Unifying Few- and Zero-Shot Egocentric Action Recognition
• Speaker diarization with session-level speaker embedding refinement using graph neural networks
• Multi-Observation Elicitation
• Adaptive Hierarchical Similarity Metric Learning with Noisy Labels
• Set2Model Networks: Learning Discriminatively To Learn Generative Models
• Constructing Graph Node Embeddings via Discrimination of Similarity Distributions