Mitigating Contextual Bias

The task of Mitigating Contextual Bias aims to address the impact of background information in datasets on classification outcomes in computer vision. The specific goal is to adjust training strategies to reduce model bias caused by background associations, thereby enhancing the model's generalization ability across different backgrounds. This task has significant application value for improving the robustness and fairness of models.