The authors of the paper tackled the problems of image classifiers by generating images that optimize a classifier-derived objective. They analyzed the behavior and decisions of image classifiers by visual counterfactual explanations, detection of systematic mistakes, and visualization of neurons to verify potential spurious features. The paper addresses the problem of systematic high-confidence predictions of classifiers, and visualization of the semantic meaning of concepts. The paper also discusses the detection of systematic failure cases of a zero-shot CLIP ImageNet classifier, and producing realistic visual counterfactuals for any image classifier.
Publication date: 30 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.17833
