Accurate and efficient post-earthquake structural assessments are critical for informed decision-making regarding repair needs and operational recovery. However, related studies, including data acquisition, damage identification, and structural condition assessment, remain confined to isolated tasks. Moreover, challenges persist in accessing high-quality training data, interpreting observed damage, and developing a unified and executable workflow. To address such challenges, this study proposes a comprehensive methodology for autonomous post-earthquake structural assessment, leveraging the concept of Bidirectional Graphics-Based Digital Twin (Bi-GBDT). The Bi-GBDT framework integrates physically grounded simulations, photorealistic graphic modeling, and data-driven learning to formalize structural condition representation, enabling both forward generation of synthetic damage environment and inverse inference from observed images. The proposed assessment workflow proceeds through four main stages: establishment of the GBDT, automated data collection, damage identification, and structural assessment. In this study, a high-fidelity digital twin of a three-story reinforced concrete shear wall building is developed, serving as a testbed simulated under 120 earthquake events to provide data with visual realism and physical accuracy. The proposed methodology achieves a mean Intersection over Union (IoU) of 90.4% for multi-class damage segmentation and an accuracy of 86.7% for structural assessment, demonstrating robust performance in bridging localized observed damage with comprehensive structural evaluations. The resulting methodology represents a significant step toward direct, automated, and explainable post-earthquake structural assessments.