Research on the Method for Improving and Evaluating the Spatial Segmentation Clarity of Plan-like View Based on pix2pixHD
In the research on generating architectural plan by algorism of image translation,there are problems of difficulty in training,difficulty in sampling,difficulty in controlling and difficulty in evaluating,etc.In order to explore more effective training and evaluating methods,based on previous research,this paper explores the application method of generating architectural"plan-like view"by classical algorithm pix2pixHD.The learning preference of algorithm is found in the sample experiment,and through combined application of pix2pixHD algorithm and"sandwich wire"sample,the learning effect of the generated model is improved,so that the space dividing line of the generated graph is clearer;in the algorithm experiment,through the calculation method that improves the similarity between the evaluation generated graph and the real image pixel,the accuracy of evaluation is improved,and the algorithm of"vector ratio"used for quantitative evaluation of spatial segmentation clarity is put forward.By taking advantage of the strong points of the two evaluation methods,the learning effect of model and the quality of generated graph are quantified,the generated model with good generating effect and certain innovation ability is obtained.The research result can preliminary verify the feasibility of the route from condition graph to vector result graph based on algorithm of image translation,and it can provide theoretical and practical basis for training the generating model for high quality architectural plan.
NETL
NSTL
万方数据
