This paper proposes a Domain Adaptive Faster R-CNN model for cross-domain object detection. The model addresses domain shift by adapting to both image-level and instance-level differences between source and target domains. The approach is based on the state-of-the-art Faster R-CNN model and incorporates two domain adaptation components: one for image-level adaptation and one for instance-level adaptation. These components are designed to minimize domain discrepancy using H-divergence theory and adversarial training. A consistency regularization is further applied to learn a domain-invariant region proposal network (RPN) in the Faster R-CNN model. The model is evaluated on multiple datasets including Cityscapes, KITTI, and SIM10K, demonstrating its effectiveness in reducing domain shift and improving object detection performance across different scenarios. The results show that the proposed approach outperforms the baseline Faster R-CNN in various domain shift scenarios, highlighting its effectiveness for cross-domain object detection. The model is trained end-to-end and can be applied to real-world applications such as autonomous driving, where domain shift is a significant challenge. The paper also discusses the theoretical foundations of the approach, including a probabilistic perspective on domain shift and the use of H-divergence for domain alignment. The model is validated through extensive experiments, showing that it achieves significant improvements in mean average precision (mAP) across different domains. The approach is particularly effective in scenarios involving synthetic data, adverse weather conditions, and cross-camera adaptation. The results demonstrate that the proposed method is robust to domain shift and can be applied to a wide range of real-world object detection tasks.This paper proposes a Domain Adaptive Faster R-CNN model for cross-domain object detection. The model addresses domain shift by adapting to both image-level and instance-level differences between source and target domains. The approach is based on the state-of-the-art Faster R-CNN model and incorporates two domain adaptation components: one for image-level adaptation and one for instance-level adaptation. These components are designed to minimize domain discrepancy using H-divergence theory and adversarial training. A consistency regularization is further applied to learn a domain-invariant region proposal network (RPN) in the Faster R-CNN model. The model is evaluated on multiple datasets including Cityscapes, KITTI, and SIM10K, demonstrating its effectiveness in reducing domain shift and improving object detection performance across different scenarios. The results show that the proposed approach outperforms the baseline Faster R-CNN in various domain shift scenarios, highlighting its effectiveness for cross-domain object detection. The model is trained end-to-end and can be applied to real-world applications such as autonomous driving, where domain shift is a significant challenge. The paper also discusses the theoretical foundations of the approach, including a probabilistic perspective on domain shift and the use of H-divergence for domain alignment. The model is validated through extensive experiments, showing that it achieves significant improvements in mean average precision (mAP) across different domains. The approach is particularly effective in scenarios involving synthetic data, adverse weather conditions, and cross-camera adaptation. The results demonstrate that the proposed method is robust to domain shift and can be applied to a wide range of real-world object detection tasks.