Precise, automated segmentation of the aorta and its branches from Computed Tomography Angiography (CTA) is crucial for preoperative planning of endovascular treatments. However, this task is challenging due to the complex topology, significant anatomical variation, and the fine-scale nature of smaller vessels. This paper presents a coarse-to-fine, two-stage segmentation framework built upon the robust nnUNet architecture to address these challenges. In the first stage, a coarse segmentation is performed on downsampled images to efficiently localize a region of interest (ROI) containing the entire aorta. The second stage employs a fine-grained segmentation network on the full-resolution cropped ROI, enabling detailed and accurate delineation of aortic branches and zones. Our method utilizes an ensemble of nnUNet models with residual encoder backbones (ResEncL and ResEncM) to enhance feature extraction and improve generalization. The final segmentation is produced by a weighted average of predictions from multiple models. On the AortaSeg24 challenge testing dataset, our method achieves a competitive mean Dice Similarity Coefficient (DSC) of 0.782 and a mean Normalized Surface Distance (NSD) of 0.817, demonstrating the effectiveness of the two-stage approach for complex vascular segmentation. The source code is publicly available at https://github.com/MaxwellEng/MICCAI_CHANLLENGE24_HJL .