Counterfactual Visual Explanation via Causally-Guided Adversarial Steering
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909810004328448 |
|---|---|
| author | Qiao, Yiran Liu, Disheng Lu, Yiren Yin, Yu Du, Mengnan Ma, Jing |
| author_facet | Qiao, Yiran Liu, Disheng Lu, Yiren Yin, Yu Du, Mengnan Ma, Jing |
| contents | Recent work on counterfactual visual explanations has contributed to making artificial intelligence models more explainable by providing visual perturbation to flip the prediction. However, these approaches neglect the causal relationships and the spurious correlations behind the image generation process, which often leads to unintended alterations in the counterfactual images and renders the explanations with limited quality. To address this challenge, we introduce a novel framework CECAS, which first leverages a causally-guided adversarial method to generate counterfactual explanations. It innovatively integrates a causal perspective to avoid unwanted perturbations on spurious factors in the counterfactuals. Extensive experiments demonstrate that our method outperforms existing state-of-the-art approaches across multiple benchmark datasets and ultimately achieves a balanced trade-off among various aspects of validity, sparsity, proximity, and realism. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_09881 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Counterfactual Visual Explanation via Causally-Guided Adversarial Steering Qiao, Yiran Liu, Disheng Lu, Yiren Yin, Yu Du, Mengnan Ma, Jing Computer Vision and Pattern Recognition Recent work on counterfactual visual explanations has contributed to making artificial intelligence models more explainable by providing visual perturbation to flip the prediction. However, these approaches neglect the causal relationships and the spurious correlations behind the image generation process, which often leads to unintended alterations in the counterfactual images and renders the explanations with limited quality. To address this challenge, we introduce a novel framework CECAS, which first leverages a causally-guided adversarial method to generate counterfactual explanations. It innovatively integrates a causal perspective to avoid unwanted perturbations on spurious factors in the counterfactuals. Extensive experiments demonstrate that our method outperforms existing state-of-the-art approaches across multiple benchmark datasets and ultimately achieves a balanced trade-off among various aspects of validity, sparsity, proximity, and realism. |
| title | Counterfactual Visual Explanation via Causally-Guided Adversarial Steering |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2507.09881 |