Counterfactual Zero-Shot and Open-Set Visual Recognition

TL;DR

This paper introduces a counterfactual framework that improves zero-shot learning and open-set recognition by generating faithful, sample-specific counterfactuals, leading to better generalization and balanced recognition of seen and unseen classes.

Contribution

The paper proposes a novel counterfactual generation method that ensures faithfulness, enhancing zero-shot and open-set recognition performance by addressing distribution mismatch issues.

Findings

Significant performance improvements on ZSL and OSR benchmarks

Effective mitigation of seen/unseen class imbalance

Framework serves as a new baseline for generalization evaluation

Abstract

We present a novel counterfactual framework for both Zero-Shot Learning (ZSL) and Open-Set Recognition (OSR), whose common challenge is generalizing to the unseen-classes by only training on the seen-classes. Our idea stems from the observation that the generated samples for unseen-classes are often out of the true distribution, which causes severe recognition rate imbalance between the seen-class (high) and unseen-class (low). We show that the key reason is that the generation is not Counterfactual Faithful, and thus we propose a faithful one, whose generation is from the sample-specific counterfactual question: What would the sample look like, if we set its class attribute to a certain class, while keeping its sample attribute unchanged? Thanks to the faithfulness, we can apply the Consistency Rule to perform unseen/seen binary classification, by asking: Would its counterfactual still look like itself? If ``yes'', the sample is from a certain class, and ``no'' otherwise. Through extensive experiments on ZSL and OSR, we demonstrate that our framework effectively mitigates the seen/unseen imbalance and hence significantly improves the overall performance. Note that this framework is orthogonal to existing methods, thus, it can serve as a new baseline to evaluate how ZSL/OSR models generalize. Codes are available at https://github.com/yue-zhongqi/gcm-cf.