Does Object Recognition Work for Everyone?

By Terrance Devries, Ishan Misra, Changhan Wang, L. V. D. van der Maaten. 2019

293 citations recorded by the source corpus.

The paper analyzes the accuracy of publicly available object-recognition systems on a geographically diverse dataset. This dataset contains household items and was designed to have a more representative geographical coverage than commonly used image datasets in object recognition. We find that the systems perform relatively poorly on household items that commonly occur in countries with a low household income. Qualitative analyses suggest the drop in performance is primarily due to appearance differences within an object class (e.g., dish soap) and due to items appearing in a different context (e.g., toothbrushes appearing outside of bathrooms). The results of our study suggest that further work is needed to make object-recognition systems work equally well for people across different countries and income levels.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety, Robustness & Interpretability

Citations in the corpus

AI and the Everything in the Whole Wide World Benchmark (2021, background): “…geo-diversity of sources manifests in poor dataset representation, as the dataset becomes visually anchored to a speciﬁc dominant cultural context. de Vries et al. [2019] found that if ImageNet was developed using the results of Hindi online image queries, rather than English ones, the result…”
Designing Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs (2021, background): “[21] compared the accuracies of six object-classification systems using images of household objects from fifty countries, finding that all six systems had substantially lower accuracies for images from lower-income countries and households; and Koenecke et al.”
Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation (2021, neutral): “For example, models trained on general object recognition datasets can be discriminatory towards lower-income households (DeVries et al., 2019).”
Flamingo: a Visual Language Model for Few-Shot Learning (2022, background): “However, current automatic vision system performance has been reported to vary with race, gender or when applied across different demographics and geographic regions [12, 21, 97].”
Segment Anything (2023, methodology): “Following [29], we retrieve geographic metadata using the Flickr API.”
Understanding and Evaluating Racial Biases in Image Captioning (2021, background): “The root causes of dataset bias are complex: they stem from bias in image search engines [52], data collection practices [74, 24], and real-world disparities.”
Scaling Autoregressive Models for Content-Rich Text-to-Image Generation (2022, background): “, wedding attire, food, etc [97, 98]) in ways that might lead to miscommunication.”
Towards fairer datasets: filtering and balancing the distribution of the people subtree in the ImageNet hierarchy (2019, background): “[19] showed that most images come from Europe and the United States, and the resulting models have difficulty generalizing to images from other places.”
Data and its (dis)contents: A survey of dataset development and use in machine learning research (2020, methodology): “Meanwhile, the images in object recognition datasets have been overwhelmingly sourced from Western countries.(14) Zhao et al.”
Do Datasets Have Politics? Disciplinary Values in Computer Vision Dataset Development (2021, neutral): “Object recognition datasets have a documented bias towards Western countries [27, 105].”
Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructure (2020, background): “Data is frequently identified as a key source of these failures through its role in “bias-laundering” [40, 51, 54, 119, 125].”
SWAD: Domain Generalization by Seeking Flat Minima (2021, background): “Even in a simple image recognition scenario, systems rely on wrong cues for their prediction, e.g. , geographic distribution [3], demographic statistics [4], texture [5], or backgrounds [6].”
On the genealogy of machine learning datasets: A critical history of ImageNet (2021, background): “For example, recent papers critiquing image classification datasets envision solutions whereby imagery is collected from around the world (DeVries et al., 2019), and in particular the developing world (Shankar et al.”
Characterising Bias in Compressed Models (2020, background): “…facial analysis datasets, object detection datasets tend to under-represent images from lower income and non-Western regions (Shankar et al., 2017; DeVries et al., 2019), activity recognition datasets exhibit stereotype-aligned gender biases (Zhao et al., 2017), and word co-occurrences within…”
Bugs in the Data: How ImageNet Misrepresents Biodiversity (2022, unknown): “…numerous studies of ImageNet from different angles, ranging from the quality of its labels (Beyer et al. 2020; Tsipras et al. 2020; Northcutt, Athalye, and Mueller 2021) to the biases that it reflects (Prabhu and Birhane 2021; Crawford and Paglen 2019; Shankar et al. 2017; De Vries et al. 2019).”
Towards Measuring Fairness in AI: The Casual Conversations Dataset (2021, background): “Introduction Fairness in AI is an emerging topic in computer vision [4, 26] and has proven indispensable to develop unbiased AI models that are fair and inclusive to individuals from any background.”
Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision (2022, methodology): “We use the DollarStreet dataset [30] and benchmark protocol [51] for evaluating the disparity in object recognition accuracy in different parts of the world.”