mlsanityData to 7 Oct 2026

DeepFace-EMD: Re-ranking Using Patch-wise Earth Mover's Distance Improves Out-Of-Distribution Face Identification

arXiv 2112.04016 ↗First posted 7 Dec 2021cs.CV cs.IR cs.LGcited by 20

Not on the map. Why?

Abstract (arXiv)

Face identification (FI) is ubiquitous and drives many high-stake decisions made by law enforcement. State-of-the-art FI approaches compare two images by taking the cosine similarity between their image embeddings. Yet, such an approach suffers from poor out-of-distribution (OOD) generalization to new types of images (e.g., when a query face is masked, cropped, or rotated) not included in the training set or the gallery. Here, we propose a re-ranking approach that compares two faces using the Earth Mover's Distance on the deep, spatial features of image patches. Our extra comparison stage explicitly examines image similarity at a fine-grained level (e.g., eyes to eyes) and is more robust to OOD perturbations and occlusions than traditional FI. Interestingly, without finetuning feature extractors, our method consistently improves the accuracy on all tested OOD queries: masked, cropped, rotated, and adversarial while obtaining similar results on in-distribution images.

Key facts. Cited by 20 papers in the map (data through 7 Oct 2026). Builds on 29 of its 78 references that are in the map. First posted on arXiv 7 Dec 2021.

Builds on

29 of 78 references.

References · 29

  1. 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,833
  2. 2013Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distancescited by 2,277
  3. 2016Identity Mappings in Deep Residual Networkscited by 1,960
  4. 2016Adversarial examples in the physical worldcited by 1,762
  5. 2016Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learningcited by 1,509
  6. 2018CosFace: Large Margin Cosine Loss for Deep Face Recognitioncited by 968
  7. 2017VGGFace2: A dataset for recognising faces across pose and agecited by 911
  8. 2016Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networkscited by 895
  9. 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 774
  10. 2017SphereFace: Deep Hypersphere Embedding for Face Recognitioncited by 733
  11. 2014Learning Face Representation from Scratchcited by 451
  12. 2020A Metric Learning Reality Checkcited by 317
17 more
  1. 2017Cross-Age LFW: A Database for Studying Cross-Age Face Recognition in Unconstrained Environmentscited by 258
  2. 2021On the Robustness of Vision Transformers to Adversarial Examplescited by 167
  3. 2020Towards Transferable Adversarial Attack against Deep Face Recognitioncited by 79
  4. 2017A Pose-Sensitive Embedding for Person Re-Identification with Expanded Cross Neighborhood Re-Rankingcited by 74
  5. 2014Deeply learned face representations are sparse, selective, and robustcited by 56
  6. 2019Occlusion Robust Face Recognition Based on Mask Learning with PairwiseDifferential Siamese Networkcited by 51
  7. 2020DeepEMD: Differentiable Earth Mover's Distance for Few-Shot Learningcited by 49
  8. 2020Masked Face Recognition for Secure Authenticationcited by 47
  9. 2021Towards Interpretable Deep Metric Learning with Structural Matchingcited by 30
  10. 2019Visualizing Deep Similarity Networkscited by 24
  11. 2021End2End Occluded Face Recognition by Masking Corrupted Featurescited by 24
  12. 2017Faster Than Real-time Facial Alignment: A 3D Spatial Transformer Network Approach in Unconstrained Posescited by 19
  13. 2016Robust LSTM-Autoencoders for Face De-Occlusion in the Wildcited by 17
  14. 2018Face Synthesis for Eyeglass-Robust Face Recognitioncited by 12
  15. 2021MLFW: A Database for Face Recognition on Masked Facescited by 10
  16. 2017Enhancing Convolutional Neural Networks for Face Recognition with Occlusion Maps and Batch Triplet Losscited by 9
  17. 2020On Improving the Generalization of Face Recognition in the Presence of Occlusionscited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 5; 2023: 6; 2024: 3; 2025: 4; 2026: 2.

20 papers.

Sort:

By area:

  1. 2022Visual correspondence-based explanations improve AI robustness and human-AI team accuracycited by 37
  2. 2022How explainable are adversarially-robust CNNs?cited by 9
  3. 2022Dynamic Feature Pruning and Consolidation for Occluded Person Re-Identificationcited by 7
  4. 2023Breaching FedMD: Image Recovery via Paired-Logits Inversion Attackcited by 7
  5. 2024Enhancing person re-identification via Uncertainty Feature Fusion Method and Auto-weighted Measure Combinationcited by 6
  6. 2023EvalAttAI: A Holistic Approach to Evaluating Attribution Maps in Robust and Non-Robust Modelscited by 5
  7. 2023Fast and Interpretable Face Identification for Out-Of-Distribution Data Using Vision Transformerscited by 4
  8. 2023PCNN: Probable-Class Nearest-Neighbor Explanations Improve Fine-Grained Image Classification Accuracy for AIs and Humanscited by 3
  9. 2022OTFace: Hard Samples Guided Optimal Transport Loss for Deep Face Representationcited by 2
  10. 2024TAB: Transformer Attention Bottlenecks enable User Intervention and Debugging in Vision-Language Modelscited by 2
  11. 2025A Re-ranking Method using K-nearest Weighted Fusion for Person Re-identificationcited by 2
  12. 2025Building Trustworthy AI for Materials Discovery: From Autonomous Laboratories to Z-scorescited by 2
8 more
  1. 2022A Survey of Face Recognitioncited by 1
  2. 2023Prototype Learning for Explainable Brain Age Predictioncited by 0
  3. 2023Bidirectional Knowledge Reconfiguration for Lightweight Point Cloud Analysiscited by 0
  4. 2024Optimisation of federated learning settings under statistical heterogeneity variationscited by 0
  5. 2025MoDE: Mixture of Diffusion Experts for Any Occluded Face Recognitioncited by 0
  6. 2025DetailSemNet: Elevating Signature Verification through Detail-Semantic Integrationcited by 0
  7. 2026MGFace: Mask-Gated Face Matching via Conditional Similarity Routingcited by 0
  8. 2026Video-Based Palm-Vein Authentication under Challenging Conditionscited by 0