Our dataset encompasses three common types of images used in the architectural design process, each paired with the corresponding point cloud: architectural line drawings (AL), architectural shading diagrams (AS), and architectural renderings (AR), each accompanied by their respective masks. These represent different stages of the design process, from the initial conception and layout to detailed design intricacies, and culminating in complete visual presentations, all necessitating a conversion from two to three dimensions. Our dataset has the following three characteristics:
High-Quality Sources: Because the designs within the dataset are completed by architecture students under professional guidance and review, each design has undergone rigorous evaluation and refinement to ensure the quality of the data.
Uniformity and Diversity: Because of the specific requirements of the project, our dataset consists of architectural masses designed by 131 students for the same rectangular site. Although all the designs are confined to the same plot, diversity was introduced through each student's distinct design thinking, style, and creativity.
Rich Perspective Information: To better represent the 3D form of the architecture, we captured images associated with each architectural model from multiple angles and recorded detailed camera parameters.Example instances from our ArchiCloudNet dataset.ArchiCloudNet 3D assets are semantically diverse, highquality, and paired with camera parameters
