mlsanityData to 14 Oct 2026

Understanding and Evaluating Racial Biases in Image Captioning

arXiv 2106.08503 ↗First posted 16 Jun 2021cs.CVcited by 108

Not on the map. Why?

Cited by 30 in Multimodal LLMs / vision-language models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Image captioning is an important task for benchmarking visual reasoning and for enabling accessibility for people with vision impairments. However, as in many machine learning settings, social biases can influence image captioning in undesirable ways. In this work, we study bias propagation pathways within image captioning, focusing specifically on the COCO dataset. Prior work has analyzed gender bias in captions using automatically-derived gender labels; here we examine racial and intersectional biases using manual annotations. Our first contribution is in annotating the perceived gender and skin color of 28,315 of the depicted people after obtaining IRB approval. Using these annotations, we compare racial biases present in both manual and automatically-generated image captions. We demonstrate differences in caption performance, sentiment, and word choice between images of lighter versus darker-skinned people. Further, we find the magnitude of these differences to be greater in modern captioning systems compared to older ones, thus leading to concerns that without proper consideration and mitigation these differences will only become increasingly prevalent. Code and data is available at https://princetonvisualai.github.io/imagecaptioning-bias .

Key facts. Cited by 108 papers in the map (data through 14 Oct 2026). Builds on 45 of its 85 references that are in the map. First posted on arXiv 16 Jun 2021.

Builds on

45 of 85 references.

References · 45

  1. 2017Attention Is All You Needcited by 60,361
  2. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  3. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  4. 2014CIDEr: Consensus-based Image Description Evaluationcited by 2,046
  5. 2016Identity Mappings in Deep Residual Networkscited by 1,965
  6. 2015Microsoft COCO Captions: Data Collection and Evaluation Servercited by 1,462
  7. 2018Datasheets for Datasetscited by 1,386
  8. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,312
  9. 2017Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answeringcited by 1,312
  10. 2020Oscar: Object-Semantics Aligned Pre-training for Vision-Language Taskscited by 1,214
  11. 2020Language (Technology) is Power: A Critical Survey of "Bias" in NLPcited by 956
  12. 2016Semantics derived automatically from language corpora contain human-like biasescited by 901
33 more
  1. 2014Show and Tell: A Neural Image Caption Generatorcited by 879
  2. 2016SPICE: Semantic Propositional Image Caption Evaluationcited by 824
  3. 2016Self-critical Sequence Training for Image Captioningcited by 547
  4. 2017Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraintscited by 420
  5. 2018AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Biascited by 319
  6. 2019Measurement and Fairnesscited by 308
  7. 2019Attention on Attention for Image Captioningcited by 289
  8. 2016Image Captioning with Semantic Attentioncited by 260
  9. 2020Data and its (dis)contents: A survey of dataset development and use in machine learning researchcited by 260
  10. 2020Large image datasets: A pyrrhic win for computer vision?cited by 237
  11. 2018Women also Snowboard: Overcoming Bias in Captioning Modelscited by 236
  12. 2018Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systemscited by 234
  13. 2016Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioningcited by 232
  14. 2018A Comprehensive Survey of Deep Learning for Image Captioningcited by 214
  15. 2019Racial Bias in Hate Speech and Abusive Language Detection Datasetscited by 209
  16. 2019Towards Fairer Datasets: Filtering and Balancing the Distribution of the People Subtree in the ImageNet Hierarchycited by 156
  17. 2019Does Object Recognition Work for Everyone?cited by 151
  18. 2020Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructurecited by 138
  19. 2020REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasetscited by 136
  20. 2019Towards a Critical Race Methodology in Algorithmic Fairnesscited by 125
  21. 2020Fair Attribute Classification through Latent Space De-biasingcited by 125
  22. 2019Lessons from Archives: Strategies for Collecting Sociocultural Data in Machine Learningcited by 116
  23. 2019Fair Generative Modeling via Weak Supervisioncited by 114
  24. 201850 Years of Test (Un)fairness: Lessons for Machine Learningcited by 101
  25. 2021A Study of Face Obfuscation in ImageNetcited by 83
  26. 2019Predictive Inequity in Object Detectioncited by 77
  27. 2018Discriminability objective for training descriptive captionscited by 67
  28. 2021Directional Bias Amplificationcited by 54
  29. 2020Mitigating Gender Bias in Captioning Systemscited by 50
  30. 2019Exposing and Correcting the Gender Bias in Image Captioning Datasets and Modelscited by 36
  31. 2021One Label, One Billion Faces: Usage and Consistency of Racial Categories in Computer Visioncited by 33
  32. 2018Fairness GANcited by 32
  33. 2016Stereotyping and Bias in the Flickr30K Datasetcited by 25

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 1; 2022: 26; 2023: 27; 2024: 26; 2025: 17; 2026: 11.

108 papers.

Sort:

By area:

  1. 2022Flamingo: a Visual Language Model for Few-Shot Learningcited by 4,386
  2. 2022GIT: A Generative Image-to-text Transformer for Vision and Languagecited by 537
  3. 2022DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Modelscited by 230
  4. 2022Vision-Language Pre-training: Basics, Recent Advances, and Future Trendscited by 117
  5. 2022A Prompt Array Keeps the Bias Away: Debiasing Vision-Language Models with Adversarial Learningcited by 104
  6. 2024SafetyPrompts: a Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safetycited by 77
  7. 2022A Systematic Study of Bias Amplificationcited by 65
  8. 2023FACET: Fairness in Computer Vision Evaluation Benchmarkcited by 58
  9. 2024Examining Gender and Racial Bias in Large Vision-Language Models Using a Novel Dataset of Parallel Imagescited by 55
  10. 2023SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examplescited by 53
  11. 2023Uncurated Image-Text Datasets: Shedding Light on Demographic Biascited by 52
  12. 2024A Decade's Battle on Dataset Bias: Are We There Yet?cited by 47
38 more
  1. 2023Inspecting the Geographical Representativeness of Images from Text-to-Image Modelscited by 44
  2. 2022Gender and Racial Bias in Visual Question Answering Datasetscited by 39
  3. 2022Adaptive Testing of Computer Vision Modelscited by 39
  4. 2022Gender Artifacts in Visual Datasetscited by 38
  5. 2022Quantifying Societal Bias Amplification in Image Captioningcited by 37
  6. 2022Measuring Representational Harms in Image Captioningcited by 36
  7. 2023Exposing and Mitigating Spurious Correlations for Cross-Modal Retrievalcited by 36
  8. 2024CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?cited by 35
  9. 2022On Learning Fairness and Accuracy on Multiple Subgroupscited by 32
  10. 2023Overwriting Pretrained Bias with Finetuning Datacited by 32
  11. 2024GenderBias-VL: Benchmarking Gender Bias in Vision Language Models via Counterfactual Probingcited by 31
  12. 2022FairGRAPE: Fairness-aware GRAdient Pruning mEthod for Face Attribute Classificationcited by 30
  13. 2022Bias Mimicking: A Simple Sampling Approach for Bias Mitigationcited by 30
  14. 2023Balancing the Picture: Debiasing Vision-Language Datasets with Synthetic Contrast Setscited by 27
  15. 2024Position: Measure Dataset Diversity, Don't Just Claim Itcited by 26
  16. 2024MultiHateClip: A Multilingual Benchmark Dataset for Hateful Video Detection on YouTube and Bilibilicited by 25
  17. 2024Would Deep Generative Models Amplify Bias in Future Models?cited by 24
  18. 2023Consensus and Subjectivity of Skin Tone Annotation for ML Fairnesscited by 22
  19. 2022Men Also Do Laundry: Multi-Attribute Bias Amplificationcited by 21
  20. 2022GSRFormer: Grounded Situation Recognition Transformer with Alternate Semantic Attention Refinementcited by 20
  21. 2023Mitigating Test-Time Bias for Fair Image Retrievalcited by 20
  22. 2023Mitigating stereotypical biases in text to image generative systemscited by 19
  23. 2025Position is Power: System Prompts as a Mechanism of Bias in Large Language Models (LLMs)cited by 19
  24. 2024A Unified Debiasing Approach for Vision-Language Models across Modalities and Taskscited by 18
  25. 2023Ethical Considerations for Responsible Data Curationcited by 17
  26. 2023Model-Agnostic Gender Debiased Image Captioningcited by 17
  27. 2023Beyond Skin Tone: A Multidimensional Measure of Apparent Skin Colorcited by 17
  28. 2024The "Colonial Impulse" of Natural Language Processing: An Audit of Bengali Sentiment Analysis Tools and Their Identity-based Biasescited by 16
  29. 2022Measuring Compositional Consistency for Video Question Answeringcited by 15
  30. 2022CODER: Coupled Diversity-Sensitive Momentum Contrastive Learning for Image-Text Retrievalcited by 13
  31. 2023Metrics for Dataset Demographic Bias: A Case Study on Facial Expression Recognitioncited by 13
  32. 2024Fairness and Bias Mitigation in Computer Vision: A Surveycited by 13
  33. 2023From Fake to Real: Pretraining on Balanced Synthetic Images to Prevent Spurious Correlations in Image Recognitioncited by 12
  34. 2023Evaluating the Fairness of Discriminative Foundation Models in Computer Visioncited by 12
  35. 2024FairDeDup: Detecting and Mitigating Vision-Language Fairness Disparities in Semantic Dataset Deduplicationcited by 11
  36. 2023Graph Neural Networks in Vision-Language Image Understanding: A Surveycited by 10
  37. 2023Dense Video Object Captioning from Disjoint Supervisioncited by 10
  38. 2023Adventures of Trustworthy Vision-Language Models: A Surveycited by 10

50 most cited of 108 All 108 (JSON)