Chasing the White Rabbit: A case study of predicting design phases of architects by training a deep neural network with sketch recognition through a digital drawing boardJessica Bielski; Burak Mete; Viktor Eisenstadt; Christoph Langenhan; Frank Petzold; Andreas Dengel; Klaus-Dieter Althoff
In: Proceedings of the 13th International Conference on Computational Creativity. International Conference on Computational Creativity (ICCC-22), 13th, June 27 - July 1, Bolzano, Italy, Pages 73-77, ISBN 978-989-54160-4-2, Association for Computational Creativity (ACC), 6/2022.
Within this paper we propose an interdisciplinary approach at the interface of computer science and architecture to predict design phases using a deep neural network, based on architects’ hand drawings. The overall goal of the metis projects is to provide architects with appropriate design step suggestions using deep learning (DL) and based on semantic information of Building Information Modeling (BIM), inspired by textual autocompletion of digital keyboards on smartphones. We describe the process of our sketch protocol study and open-source software prototype developed for sketch data acquisition with a WACOM tablet and video recordings, as well as the evaluation of the sketch protocol study and the results of the recurrent neural network (RNN) with Long Short-Term Memory (LSTM) architecture, trained with the sketch data quantified through the prototype tool. The initial prediction results of the current and the consecutive design phase appear promising to predict with high accuracy. Our futureplans include tracking the architects design process through the labyrinth of design decision making using different mental layers (e.g. design phases) as filters all the way to the bottom to isolate the individual mental process of a singular design step.
Metis-II - Knowledge-based search and query methods for the development of semantic information models for use in early design phases