Noise or Signal: The Role of Image Backgrounds in Object Recognition

Abstract

We assess the tendency of state-of-the-art object recognition models to depend on signals from image backgrounds. We create a toolkit for disentangling foreground and background signal on ImageNet images, and find that (a) models can achieve non-trivial accuracy by relying on the background alone, (b) models often misclassify images even in the presence of correctly classified foregrounds--up to 87.5% of the time with adversarially chosen backgrounds, and (c) more accurate models tend to depend on backgrounds less. Our analysis of backgrounds brings us closer to understanding which correlations machine learning models use, and how they determine models' out of distribution performance.

Cited by in corpus (28)

• On the Opportunities and Risks of Foundation Models
• WILDS: A Benchmark of in-the-Wild Distribution Shifts
• Rethinking Spatial Dimensions of Vision Transformers
• Long-Short Transformer: Efficient Transformers for Language and Vision
• Reliability and Interpretability in Science and Deep Learning
• Counterfactual Generative Networks
• IAIA-BL: A Case-based Interpretable Deep Learning Model for Classification of Mass Lesions in Digital Mammography
• 3DB: A Framework for Debugging Computer Vision Models
• Time for a Background Check! Uncovering the impact of Background Features on Deep Neural Networks
• Delving Deep into the Generalization of Vision Transformers under Distribution Shifts
• Model-Based Domain Generalization
• Temporal Shuffling for Defending Deep Action Recognition Models against Adversarial Attacks
• Selective Classification Can Magnify Disparities Across Groups
• CLCC: Contrastive Learning for Color Constancy
• Re-labeling ImageNet: from Single to Multi-Labels, from Global to Localized Labels
• A Fine-Grained Analysis on Distribution Shift
• Leveraging Sparse Linear Layers for Debuggable Deep Networks
• Removing Spurious Features can Hurt Accuracy and Affect Groups Disproportionately
• Exploring Adversarial Robustness of Deep Metric Learning
• Contemplating real-world object classification
• Robust Representation Learning via Perceptual Similarity Metrics
• Object-Aware Cropping for Self-Supervised Learning
• FOCUS: Familiar Objects in Common and Uncommon Settings
• Towards Robust Classification Model by Counterfactual and Invariant Data Generation
• Excess Capacity and Backdoor Poisoning
• Adversarial Training for EM Classification Networks
• Combining Diverse Feature Priors
• Human Imperceptible Attacks and Applications to Improve Fairness