Point clouds play a pivotal role in three-dimensional perception tasks, such as virtual reality and robotics. Nevertheless, the point clouds acquired by 3D sensors frequently exhibit missing areas or fragmented structures caused by issues like specular reflection and object occlusion. Consequently, the recovery of complete point clouds from these defective ones has garnered increasing attention. Existing approaches have struggled to simultaneously preserve fine-grained details and global structural integrity in point cloud completion. To address this, we present Detail Aware CompletionNet, a novel architecture that encodes local point cloud patches into sequential representation by integrating hierarchical structural features and positional embeddings using self-attention. Subsequently, these sequential features are mapped into completed coarse features through a geometry-aware Transformer. Additionally, to generate point clouds with more comprehensive structures, we employ a multi-stage FoldingNet strategy, refining the prediction of missing point clouds from coarse to fine levels. Experiments conducted on multiple datasets confirms the effectiveness of our approach, which surpass current methods in retaining detailed structures while ensuring accurate global shapes.