A Bayesian Approach to OOD Robustness in Image Classification

2024-03-12

Computer Vision and Pattern Recognition

AI Summary

Addressing real-world scenarios where target-domain labels are unavailable and out-of-distribution (OOD) interference and occlusions coexist, this paper proposes Unsupervised Generative Transition (UGT) to enhance cross-domain generalization and occlusion robustness of image classifiers. UGT constructs a source–target domain transferable part dictionary using the von Mises–Fisher (vMF) kernel, and jointly optimizes inter-domain semantic alignment and robust local part representation via compositional neural networks, Bayesian generative modeling, and iterative dictionary refinement. Evaluated on multiple benchmarks—including OOD-CV, ImageNet-C, and synthetic-to-real domain adaptation—UGT consistently outperforms existing unsupervised methods across OOD detection, classification accuracy, and occlusion robustness. Notably, it is the first unsupervised approach to jointly model distributional shift and structural occlusion, establishing a new state of the art in unsupervised robust domain adaptation.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

An important and unsolved problem in computer vision is to ensure that the algorithms are robust to changes in image domains. We address this problem in the scenario where we only have access to images from the target domains. Motivated by the challenges of the OOD-CV [45] benchmark where we encounter real world Out-of-Domain (OOD) nuisances and occlusion, we introduce a novel Bayesian approach to OOD robustness for object classification. Our work extends Compositional Neural Networks (CompNets), which have been shown to be robust to occlusion but degrade badly when tested on OOD data. We exploit the fact that CompNets contain a generative head defined over feature vectors represented by von Mises-Fisher (vMF) kernels, which correspond roughly to object parts, and can be learned without supervision. We obverse that some vMF kernels are similar between different domains, while others are not. This enables us to learn a transitional dictionary of vMF kernels that are intermediate between the source and target domains and train the generative model on this dictionary using the annotations on the source domain, followed by iterative refinement. This approach, termed Unsupervised Generative Transition (UGT), performs very well in OOD scenarios even when occlusion is present. UGT is evaluated on different OOD benchmarks including the OOD-CV dataset, several popular datasets (e.g., ImageNet-C [9]), artificial image corruptions (including adding occluders), and synthetic-to-real domain transfer, and does well in all scenarios.

Problem

• Ensuring robustness to image domain changes
• Handling unannotated target domain images
• Improving OOD object classification accuracy

Innovation

• Bayesian approach for OOD robustness
• Unsupervised Generative Transition (UGT)
• vMF kernels for domain adaptation