Entropy guided adversarial model for weakly supervised object localization

Weakly Supervised Object Localization is challenging due to the lack of bounding box annotations. Previous works tend generate a Class Activation Map (CAM) localize object. However, CAM highlights only most discirminative part object and does not highlight whole To address this problem, we propose an Entropy Guided Adversarial model (EGA model) perform better localization objects. EGA uses adversarial learning method create examples, i.e., images where perturbation added. Treating examples as data augmentation regularize our well detect more discriminative visual pattern on CAM. We further apply Shannon entropy generated guide during training. Minimizing loss forces high-confident The detects while excludes background. Extensive experiments show that improves classification performances state-of-the-art benchmarks. Ablation also both contribute algorithm performance.