基于ControlNet的儿童床设计效果图生成技术

Children bed design effect diagram based on ControlNet

基于深度学习的计算机生成技术,可实现产品设计效果图快速生成,对提升产品设计效率、降低人工劳动强度具有积极作用,也为高质量的图像生成提供了新思路。本研究提出了一种基于ControlNet的儿童床设计效果图生成方法。首先,采用Canny边缘检测算法对原始图像进行处理,得到对应的边缘图像;然后,利用控制分支网络从边缘图像中提取特征;最后,将提取的边缘特征融合到原有的扩散模型特征层中,并作为条件来控制ControlNet的生成过程,通过设置不同的随机数种子,可生成具有相同结构但不同配色的目标图像。以FID值、Inception Score值为指标,对试验结果进行定量评估,结果表明:相较于CycleGAN、ChipGAN、Style2Paints 3种图像生成模型,基于ControlNet生成的图像不仅拥有较好的多样性,其图像细节质量也有所提高,证明了所应用模型的有效性和合理性。同时,对基于ControlNet的儿童床设计效果图方法进行了专家评估,评估满意度达92.22%,进一步证明该方法的可靠性和适用性。将ControlNet应用于儿童床设计效果图工作是可行的,不仅可为设计师提供设计灵感,也可节省人力与物力成本,提高设计效率。

Product design renderings have an important impact on designers plans and final product presentation.Therefore,the rapid generation of high-quality renderings for improving design efficiency and promoting the development of the design industry is of great significance.The traditional rendering method of product design renderings needs to configure the elements of 3D model,such as material,lighting and background,and use complex algorithms to obtain realistic renderings,the resulting renderings have some problems in terms of light and shadow,and often cannot express the lighting,material,and texture details of the real world perfectly.Therefore,in order to solve the problems of traditional rendering methods of product design renderings,such as time-consuming,high cost and low quality,deep learning-based image generation technology has become one of the development directions of the furniture industry.The development of deep learning technology provides a new idea for high-quality image generation,and the rapid generation of product design renderings through computer generation technology can further improve the efficiency of product design,reduce the intensity of manual labor.In this study,a method of generating children bed design effect graph using ControlNet,which is a neural network computing platform,is proposed.Firstly,Canny edge detection algorithm is used to process the original images and obtain the corresponding edge images.Then,the feature is extracted from the edge image using the control branch network.Finally,the extracted edge features are fused into the feature layer of the original diffusion model and used as a condition to control the generating process of ControlNet.By setting different seeds of random number,the target image with the same structure but different color matching can be generated.Based on Fréchet Inception Distance(FID)and Inception Score,the results show that the images generated by ControlNet not only have better diversity than those generated by CycleGan,ChipGan and Style2Paints,the image detail quality is also improved,which proves the validity and rationality of the applied model.At the same time,the effect diagram method of children bed design based on ControlNet is evaluated by experts,and the result proves the reliability and applicability of the method.It is feasible to apply ControlNet to children bed design,which could not only provide inspiration for designers,but also save human and material cost and improve design efficiency.