Undoing the Damage of Dataset Bias

By Aditya Khosla, Tinghui Zhou, Tomasz Malisiewicz, Alexei A. Efros, Antonio Torralba. 2012

579 citations recorded by the source corpus.

The presence of bias in existing object recognition datasets is now well-known in the computer vision community. While it remains in question whether creating an unbiased dataset is possible given limited resources, in this work we propose a discriminative framework that directly exploits dataset bias during training. In particular, our model learns two sets of weights: (1) bias vectors associated with each individual dataset, and (2) visual world weights that are common to all datasets, which are learned by undoing the associated bias from each dataset. The visual world weights are expected to be our best possible approximation to the object model trained on an unbiased dataset, and thus tend to have good generalization ability. We demonstrate the effectiveness of our model by applying the learned weights to a novel, unseen dataset, and report superior results for both classification and detection tasks compared to a classical SVM that does not account for the presence of bias. Overall, we find that it is beneficial to explicitly account for bias when combining multiple datasets.

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

Webly Supervised Joint Embedding for Cross-Modal Image-Text Retrieval (2018, background): “Moreover, there exists different types of bias in the existing datasets [28, 49, 50].”
Predicting Deep Zero-Shot Convolutional Neural Networks Using Textual Descriptions (2015, background): “In [12] the authors explicitly account for inter-dataset biases and are able to train a model that is invariant to these.”
Multiple-Source Adaptation with Domain Classifiers (2020, background): “Khosla et al. [2012], Blanchard et al. [2011] combined all the source data to jointly train a single predictor.”
Algorithms and Theory for Multiple-Source Adaptation (2018, methodology): “We used the AlexNet Krizhevsky et al. (2012) ConvNet (CNN) architecture, and used the output from the softmax score layer as our base predictors, pre-trained on ImageNet and used fc7 activations as our features for density estimation Donahue et al.”
Diversity in Faces (2019, methodology): “as a domain transfer problem. Tommasi et al. [47] and Homan et al. [48] conducted a similar evaluation with deep features, showing that data set bias can be reduced but not eliminated. Khosla et al. [49] proposed a method that learns bias vectors associated with individual data sets, as well as weig”
Taking Advantage of Multitask Learning for Fair Classification (2018, neutral): “These results are in line with previous studies on MTL, which suggest the benefit offered by this methodology, see [15 18 27] and references therein.”
Domain Generalization via Conditional Invariant Representations (2018, methodology): “The two real-world image classiﬁ-cation datasets are the VLCS and Ofﬁce+Caltech datasets, which are widely used datasets to evaluate the performance of domain generalization and domain adaptation (Ghifary et al. 2017; Gong et al. 2016; Khosla et al. 2012).”
Deep Domain Confusion: Maximizing for Domain Invariance (2014, background): “In the last few years several methods for visual domain adaptation have been suggested to overcome this issue [10, 33, 2, 29, 25, 22, 17, 16, 19, 20], but were limited to shallow models.”
In Search of Lost Domain Generalization (2020, background): “Li et al. [2017] extends Khosla et al. [2012] to deep neural networks by extending each of their parameter tensors with one additional dimension, indexed by the training domains, and set to a neutral value to predict domain-agnostic test examples.”
Deeper, Broader and Artier Domain Generalization (2017, positive): “Undo-Bias: Modifies traditional SVM to include a domain-specific and global weight vector which can be extracted for DG [12].”
One-Shot Adaptation of Supervised Deep Convolutional Models (2013, methodology): “Many visual domain adaptation methods have been put forth to compensate for dataset bias [7, 22, 1, 20, 18, 16, 13, 12, 14, 15], but are limited to shallow models.”
Class-Incremental Domain Adaptation (2020, background): “In the presence of a larger domain-shift [43] between the training and the test domains, the performance of deep models degrades drastically resulting from the domain-bias [18,50].”
Unified Deep Supervised Domain Adaptation and Generalization (2017, background): “Specifically, [32] adjusts the weights of the classifier to work well on an unseen dataset, and [65] fuses the score of exemplar classifiers given any test sample.”
A Deeper Look at Dataset Bias (2015, methodology): “We focus here on the method proposed in [25] to overcome the dataset bias.”
No More Discrimination: Cross City Adaptation of Road Scene Segmenters (2017, methodology): “Inspired by the recent advances in domain adaptation [23, 35, 12], we propose an unsupervised learning framework for performing cross-city semantic segmentation.”
Scene Text Visual Question Answering (2019, background): “A key benefit of combining images from various datasets is the reduction of dataset bias such as selection, capture and negative set bias which have been shown to exist in popular image datasets[30].”