基于pix2pixHD的类平面图空间分割清晰度提升及评价方法研究  

Research on the Method for Improving and Evaluating the Spatial Segmentation Clarity of Plan-like View Based on pix2pixHD

作　　者：崔哲[1] 郭昱 森敏惠 李华[1] CUI Zhe;GUO Yu;SEN Minhui;LI Hua

机构地区：[1]同济大学建筑与城市规划学院

出　　处：《住宅科技》2024年第8期20-26,共7页Housing Science

摘　　要：用图像转译算法生成建筑平面图的研究存在训练难、取样难、控制难、评价难等问题。为了探索更有效的训练及评价方法,文章基于先前研究,探索经典算法pix2pixHD在建筑“类平面图”生成上的应用方法。在样本实验中,发现算法的学习偏好,通过对pix2pixHD算法与“夹芯线”样本的结合运用,提升生成模型的学习效果,使生成图的空间分割线更加清晰;在算法实验中,通过改进评价生成图与真实图像素相似度的计算方法,提升评价的精确度,并提出用于量化评价空间分割清晰度的“矢量比值”算法。通过两种评价方法的优势互补,量化模型的学习效果和生成图质量,以获得生成效果好且有一定创新能力的生成模型。研究结果可初步验证基于图像转译算法,从条件图到矢量结果图路线的可行性,可为训练高质量的建筑平面图生成模型提供理论与实践基础。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.

关 键 词：数字化 图像转译 类平面图 空间分割清晰度 生成模型 评价方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]