Realistic and delicate window models can be applied to virtual reality, urban planning and other fields, but the current re-search is too simplified to describe complex window types and structural parameters. This paper presents a reconstruction method of window procedural model based on sketch, which defines many window types in advance and creates corresponding grammar parame-ters for them. Users only need to draw window sketch as input and the system can quickly identify and estimate the grammar parame-ters of windows and automatically generate the final 3D model by relying on the powerful recognition ability of convolutional neural net-work. This paper uses non-realistic rendering technology to generate rich training datasets, and based on attention mechanism, can accurately identify grammar parameters. Experimental results verify the effectiveness and accuracy of the method.
维普
万方数据
