The Household Design Based On Deep Neural Network

The household design is an important direction for indoor scene understanding.A reasonable and effective intelligent home model can provide a lot of help for design work and has a good application prospect.Currently home layout methods are all based on traditional machine learning algorithms and require multiple models to complete tasks together.Therefore,there is room for improvement in model simplification.At the same time,in the field of art design,spatial sequence level household design is often an important way to master the overall space and control design methods.Therefore,inspired by the idea of sequence,this paper proposes a new household design method based on deep neural network,which has a certain significance for the application of deep learning technology.Firstly,in order to verify the validity of the sequence idea,this paper proposes a sequential structure household design method.In the data preprocessing stage,we annotate the data sets needed for this model,and use the Encoder-Decoder framework based on the recurrent neural network and the sequence model proposed in recent years to achieve the sequential structure prediction task.Given a household sequence,this model will extract temporal characteristics between the characters and predict the next household sequence,linking the given sequence of conditions with the target prediction to form a complete home design solution.Secondly,we propose a bidirectional hierarchical household design model.In order to make the model closer to the actual application,we introduce parameter constraints,that is,limit the relative size of household objects,using the improved model on the EncoderDecoder to build a bi-directional layered household design model and achieve household serial design work.The complete bidirectional hierarchical household design is to divide the prediction task into four modules based on four walls,including data preprocessing,feature extraction,hierarchical prediction,and parameter screening.The hierarchical prediction and parameter selection are two modules that are performed at the same time,that is,for each layer of the prediction task,we must give the parameter constraints of the layer in advance,and then filter the parameters in the prediction stage.Finally,we use the Unity3 D engine to render the household scene designed by our model.By comparing with the popular machine learning methods in recent years,we have shown that our model has a certain reference value.Meanwhile,we tried to learn the style of the household and it shows that the sequence thinking based on deep learning has certain theoretical significance and practical value in the field of household design.In the end,we have summarized and looked forward to the work of this paper.Then,we analyzed some problems and it is the direction that this work should refer to and improve in the future.