New progress has been made in the field of computer -aided design in Hefei University of Technology

Recently, the joint research team of the School of Computer, Information College and Software of Hefei University of Technology has achieved important research progress in the field of computer -aided design. Transactions on Graphics (Tog). This is the first time that the school has published papers in the above -mentioned international conferences and journals.

With the rapid development of computer industry technology, indoor scenes are widely used in the fields of animation films, computer games, virtual reality, etc., and are important research objects in computer graphics and computer vision. The indoor layout floor plan is the basis for indoor design and scene modeling. The traditional layout floor map is manually created by the designer or assisted by software. The design process is tedious and lengthy. The use of artificial intelligence technology to learn and automatically generate indoor layout floor plans are of great significance for architecture, computer graphics and other fields.

In recent years, many progress has been made based on the plane layout generation method based on deep learning, but to obtain the ideal results, complicated post -treatment is still needed. In order to achieve a more thorough automatic generating result, the research team has designed the Wall Graph data structure for layout and modeling. With the support of the public data set RPLAN (including more than 80,000 real building interior design drawings), it is based on The method of deep learning directly synthesizes Wall Graph to generate the indoor layout plan. In order to achieve this purpose, the research team integrates the traversal of the map to the deep neural network, designed the geometric network and semantic network, and coupled the two deep neural networks to generate WALL Graph by the return to iterative iterations.

Compared with the previous research results, the work was for the first time to achieve automatic generating end -to -end generation, as well as more constraints, including the constraints of learning bearing -based load -bearing walls, and achieved leading results. The paper was invited to give a lecture report at the SIGGRAPH 2022 conference held in Canada from August 8th to 11th this year.

The first author of the paper is Sun Jiahui, a master's degree in the School of Computer and Information, and the corresponding author is Professor Zheng Liping of the School Software Academy, Computer and Information College and Dr. Wu Civilization. The school is the first completion unit of the paper. Professor Liu Ligang of the School of Mathematics and Technology of China University of Science and Technology participated in the cooperation.

This work is supported by projects such as the National Natural Science Foundation of China and the basic scientific research fund of the central university. (Correspondent: Hefei University of Technology Wu Civilization)

- END -