一种基于生成对抗网络的三维目标生成方法

A 3D object generation method based on GAN

文章提出了一种基于生成对抗网络(Generative Adversarial Networks, GAN)的三维目标生成算法,能从符合高斯正态分布的随机数中生成目标的三维图像。首先,将随机噪声输入生成器,经过一系列的卷积和Transformer块运算后,生成器生成对应目标的三维图像,然后判别器对生成的目标图像和真实的目标图像进行二值判别。通过生成器和判别器的交替训练,不断提升生成器生成三维目标的质量,最终达到准确的生成性能。

This paper proposes a target generation algorithm based on Generative Adversarial Networks(GAN),which can generate a 3D image of a target from random numbers conforming to a Gaussian normal distribution.First,random noise is input into the generator,and after a series of convolution and Transformer block operations,the generator generates a three-dimensional image corresponding to the target.Then the discriminator performs binary discrimination between the generated target image and the real target image.Through the alternate training of generator and discriminator,the quality of 3D objects generated by the generator is continuously improved,and the accurate generation performance is finally achieved.