mlsanityData to 7 Oct 2026

Optimizing Relevance Maps of Vision Transformers Improves Robustness

arXiv 2206.01161 ↗First posted 2 Jun 2022cs.CVcited by 41

Not on the map. Why?

Abstract (arXiv)

It has been observed that visual classification models often rely mostly on the image background, neglecting the foreground, which hurts their robustness to distribution changes. To alleviate this shortcoming, we propose to monitor the model's relevancy signal and manipulate it such that the model is focused on the foreground object. This is done as a finetuning step, involving relatively few samples consisting of pairs of images and their associated foreground masks. Specifically, we encourage the model's relevancy map (i) to assign lower relevance to background regions, (ii) to consider as much information as possible from the foreground, and (iii) we encourage the decisions to have high confidence. When applied to Vision Transformer (ViT) models, a marked improvement in robustness to domain shifts is observed. Moreover, the foreground masks can be obtained automatically, from a self-supervised variant of the ViT model itself; therefore no additional supervision is required.

Key facts. Cited by 41 papers in the map (data through 7 Oct 2026). Builds on 55 of its 83 references that are in the map. First posted on arXiv 2 Jun 2022.

Builds on

55 of 83 references.

References · 55

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  3. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  4. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,893
  5. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,690
  6. 2017A Unified Approach to Interpreting Model Predictionscited by 6,488
  7. 2020Training data-efficient image transformers & distillation through attentioncited by 4,560
  8. 2014Going Deeper with Convolutionscited by 4,231
  9. 2015Domain-Adversarial Training of Neural Networkscited by 3,023
  10. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  11. 2020Shortcut Learning in Deep Neural Networkscited by 1,928
  12. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,842
38 more
  1. 2019Invariant Risk Minimizationcited by 1,832
  2. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
  3. 2017Learning Important Features Through Propagating Activation Differencescited by 1,454
  4. 2019Natural Adversarial Examplescited by 1,401
  5. 2014How transferable are features in deep neural networks?cited by 1,216
  6. 2017SmoothGrad: removing noise by adding noisecited by 1,097
  7. 2019Learning Robust Global Representations by Penalizing Local Predictive Powercited by 1,010
  8. 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 950
  9. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 844
  10. 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 774
  11. 2020Quantifying Attention Flow in Transformerscited by 759
  12. 2018Deformable ConvNets v2: More Deformable, Better Resultscited by 666
  13. 2020Transformer Interpretability Beyond Attention Visualizationcited by 620
  14. 2021How to train your ViT? Data, Augmentation, and Regularization in Vision Transformerscited by 562
  15. 2015Recent Advances in Convolutional Neural Networkscited by 360
  16. 2019Unmasking Clever Hans Predictors and Assessing What Machines Really Learncited by 359
  17. 2021Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformerscited by 316
  18. 2017Right for the Right Reasons: Training Differentiable Models by Constraining their Explanationscited by 293
  19. 2018Learning Not to Learn: Training Deep Neural Networks with Biased Datacited by 255
  20. 2015Explaining NonLinear Classification Decisions with Deep Taylor Decompositioncited by 248
  21. 2017Real Time Image Saliency for Black Box Classifierscited by 246
  22. 2018Women also Snowboard: Overcoming Bias in Captioning Modelscited by 236
  23. 2019Counterfactual Data Augmentation for Mitigating Gender Stereotypes in Languages with Rich Morphologycited by 200
  24. 2019Full-Gradient Representation for Neural Network Visualizationcited by 189
  25. 2021Benchmarking Representation Learning for Natural World Image Collectionscited by 167
  26. 2022Self-Supervised Transformers for Unsupervised Object Discovery using Normalized Cutcited by 166
  27. 2020Counterfactual Samples Synthesizing for Robust Visual Question Answeringcited by 153
  28. 2018Pathologies of Neural Models Make Interpretations Difficultcited by 140
  29. 2020On Robustness and Transferability of Convolutional Neural Networkscited by 98
  30. 2021Large-scale Unsupervised Semantic Segmentationcited by 98
  31. 2015Visualizing Deep Convolutional Neural Networks Using Natural Pre-Imagescited by 93
  32. 2020Don't Judge an Object by Its Context: Learning to Overcome Contextual Biascited by 91
  33. 2022Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervisioncited by 80
  34. 2022The Effects of Regularization and Data Augmentation are Class Dependentcited by 67
  35. 2020Permuted AdaIN: Reducing the Bias Towards Global Statistics in Image Classificationcited by 63
  36. 2018Not Using the Car to See the Sidewalk: Quantifying and Controlling the Effects of Context in Classification and Segmentationcited by 49
  37. 2019Relative Attributing Propagation: Interpreting the Comparative Contributions of Individual Units in Deep Neural Networkscited by 49
  38. 2022Rethinking Attention-Model Explainability through Faithfulness Violation Testcited by 43

50 most cited of 55 All 55 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 3; 2023: 9; 2024: 16; 2025: 8; 2026: 5.

41 papers.

Sort:

By area:

  1. 2023Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Modelscited by 696
  2. 2022Semantic Abstraction: Open-World 3D Scene Understanding from 2D Vision-Language Modelscited by 95
  3. 2024The Hidden Attention of Mamba Modelscited by 88
  4. 2022A Whac-A-Mole Dilemma: Shortcuts Come in Multiples Where Mitigating One Amplifies Otherscited by 85
  5. 2023ImageNet-Hard: The Hardest Images Remaining from a Study of the Power of Zoom and Spatial Biases in Image Classificationcited by 33
  6. 2023Channel Vision Transformers: An Image Is Worth 1 x 16 x 16 Wordscited by 33
  7. 2024Backward Lens: Projecting Language Model Gradients into the Vocabulary Spacecited by 30
  8. 2024Parameter Efficient Fine-tuning of Self-supervised ViTs without Catastrophic Forgettingcited by 17
  9. 2023Studying How to Efficiently and Effectively Guide Models with Explanationscited by 16
  10. 2024Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulationcited by 11
  11. 2023Contextual Vision Transformers for Robust Representation Learningcited by 10
  12. 2023The Role of Masking for Efficient Supervised Knowledge Distillation of Vision Transformerscited by 8
29 more
  1. 2024Tilt your Head: Activating the Hidden Spatial-Invariance of Classifierscited by 7
  2. 2024ACTRESS: Active Retraining for Semi-supervised Visual Groundingcited by 7
  3. 2024Attention-aware Semantic Communications for Collaborative Inferencecited by 6
  4. 2024Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Trainingcited by 6
  5. 2024GenFormer -- Generated Images are All You Need to Improve Robustness of Transformers on Small Datasetscited by 6
  6. 2025Comprehensive Information Bottleneck for Unveiling Universal Attribution to Interpret Vision Transformerscited by 6
  7. 2025FOCUS: Fused Observation of Channels for Unveiling Spectracited by 6
  8. 2024LibraGrad: Balancing Gradient Flow for Universally Better Vision Transformer Attributionscited by 5
  9. 2023Neglected Free Lunch -- Learning Image Classifiers Using Annotation Byproductscited by 4
  10. 2025SpiLiFormer: Enhancing Spiking Transformers with Lateral Inhibitioncited by 3
  11. 2025A Meaningful Perturbation Metric for Evaluating Explainability Methodscited by 3
  12. 2025X-Edit: Detecting and Localizing Edits in Images Altered by Text-Guided Diffusion Modelscited by 3
  13. 2023PV2TEA: Patching Visual Modality to Textual-Established Information Extractioncited by 2
  14. 2024MaskInversion: Localized Embeddings via Optimization of Explainability Mapscited by 2
  15. 2024SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformerscited by 2
  16. 2024FastRM: An efficient and automatic explainability framework for multimodal generative modelscited by 2
  17. 2025Kolmogorov-Arnold Attention: Is Learnable Attention Better For Vision Transformers?cited by 2
  18. 2025C-LEAD: Contrastive Learning for Enhanced Adversarial Defensecited by 2
  19. 2026Concept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustnesscited by 2
  20. 2022On Background Bias in Deep Metric Learningcited by 1
  21. 2023Benchmarking Low-Shot Robustness to Natural Distribution Shiftscited by 1
  22. 2024Explanation Regularisation through the Lens of Attributionscited by 1
  23. 2024MePT: Multi-Representation Guided Prompt Tuning for Vision-Language Modelcited by 1
  24. 2024Approximate Nullspace Augmented Finetuning for Robust Vision Transformerscited by 0
  25. 2025Machine Learning from Explanationscited by 0
  26. 2026The Confusion is Real: GRAPHIC -- A Network Science Approach to Confusion Matrices in Deep Learningcited by 0
  27. 2026Shortcut Mitigation via Spurious-Positive Samplescited by 0
  28. 2026Zero-Shot Test-Time Canonicalization using Out-of-Distribution Scoringcited by 0
  29. 2026All modalities are equal, but video is more equal: Closing the Cross-Attention Gap in Joint Video Generationcited by 0