LAA-Net is a novel deepfake detection method that focuses on localized artifacts to improve detection accuracy and generalization. The paper introduces a multi-task learning framework with an explicit attention mechanism and an Enhanced Feature Pyramid Network (E-FPN) to capture low-level features effectively. The explicit attention mechanism combines heatmap-based and self-consistency strategies to focus on vulnerable pixels, which are more likely to show blending artifacts. The E-FPN helps propagate discriminative features while reducing redundancy. Experiments on multiple benchmarks show that LAA-Net outperforms existing methods in terms of AUC and AP. The method is robust to different quality of deepfakes and is effective in detecting high-quality deepfakes. The paper also includes an ablation study and comparison with traditional FPN, demonstrating the effectiveness of E-FPN in capturing localized features. The results show that LAA-Net achieves state-of-the-art performance in deepfake detection.LAA-Net is a novel deepfake detection method that focuses on localized artifacts to improve detection accuracy and generalization. The paper introduces a multi-task learning framework with an explicit attention mechanism and an Enhanced Feature Pyramid Network (E-FPN) to capture low-level features effectively. The explicit attention mechanism combines heatmap-based and self-consistency strategies to focus on vulnerable pixels, which are more likely to show blending artifacts. The E-FPN helps propagate discriminative features while reducing redundancy. Experiments on multiple benchmarks show that LAA-Net outperforms existing methods in terms of AUC and AP. The method is robust to different quality of deepfakes and is effective in detecting high-quality deepfakes. The paper also includes an ablation study and comparison with traditional FPN, demonstrating the effectiveness of E-FPN in capturing localized features. The results show that LAA-Net achieves state-of-the-art performance in deepfake detection.