Lu Xun Academy of Fine Arts, Liaoning, Dalian 116021, China.
Comput Intell Neurosci. 2022 Jun 27;2022:9187244. doi: 10.1155/2022/9187244. eCollection 2022.
The development of virtual reality technology has promoted the unceasing reform and development in the field of fashion design. Aiming at the key technologies and research difficulties in 3D clothing design, the structure of convolution neural network was designed, and a 3D clothing design model based on convolution neural network and virtual reality was constructed. By designing experiments, the average error, matching rate, prediction accuracy, pressure value, and other evaluation indexes are used to measure the performance of convolution neural network model in virtual reality three-dimensional clothing design. The results show that: first, the convolution neural network has the fastest training speed, with the maximum error of 1.63% and the average error of 0.48%. Second, with the increase of parameters, the matching degree of each part of the version and the corresponding data of the human body will gradually increase, the matching rate will slowly improve, and finally tends to 0.9 or so to achieve stability, Thirdly, the prediction accuracy of the convolution neural network is 92.689%, and the loss value is the least. Fourthly, in the virtual fitting and garment wearing experiments, the fit of the sample is within the allowed range, indicating that the automatic generation of fit sample has wearability and rationality.
虚拟现实技术的发展推动了服装设计领域的不断改革和发展。针对 3D 服装设计的关键技术和研究难点，设计卷积神经网络结构，构建基于卷积神经网络和虚拟现实的 3D 服装设计模型。通过设计实验，利用平均误差、匹配率、预测准确率、压力值等评价指标来衡量卷积神经网络模型在虚拟现实三维服装设计中的性能。结果表明：一是卷积神经网络具有最快的训练速度，最大误差为 1.63%，平均误差为 0.48%。二是随着参数的增加，版样各部分与人体对应数据的匹配度会逐渐提高，匹配率会缓慢提高，最终趋于 0.9 左右达到稳定。三是卷积神经网络的预测准确率为 92.689%，损失值最小。四是在虚拟试穿和着装实验中，样衣的合身度在允许范围内，表明合身样衣的自动生成具有可穿性和合理性。
