The Effect of Latent Space Dimension on the Quality of Synthesized Human Face Images

Abstract

In recent years Generative Adversarial Networks (GANs) have achieved remarkable results in the task of realistic image synthesis. Despite their continued success and advances, there still lacks a thorough understanding of how precisely GANs map random latent vectors to realistic-looking images and how the priors set on the latent space affect the learned mapping. In this work, we analyze the effect of the chosen latent dimension on the final quality of synthesized images of human faces and learned data representations. We show that GANs can generate images plausibly even with latent dimensions significantly smaller than the standard dimensions like 100 or 512. Although one might expect that larger latent dimensions encourage the generation of more diverse and enhanced quality images, we show that an increase of latent dimension after some point does not lead to visible improvements in perceptual image quality nor in quantitative estimates of its generalization abilities.

Generative deep learning for the development of a type 1 diabetes simulator

“…The choice of latent space was a normal distribution in our implementation. As evident from the literature, the quality of generated data of a GAN depends on the dimensions of the latent space 62 , 63 . Even though there are no standards for latent space dimensions, a size of 100 or 512 is preferred in image generation tasks.…”

Automatic generation of artificial images of leukocytes and leukemic cells using generative adversarial networks (syntheticcellgan)

“…2.B. The input is a random noise vector of size 1×1×100 with values between 0 and 1, which is a usual standard [42]. The vector is reshaped as a concatenation into an array of size 4 × 4 × 100.…”

Augmentasi Data Menggunakan Dcgan Pada Gambar Tanah

“…Gambar yang dihasilkan oleh jaringan generator merupakan gambar fotorealistik baru yang ditransformasikan berdasarkan pemetaan latent vector yang teracak dan berada pada ruang latent n-dimensi. [6]. Transformasi gambar yang bermakna dihasilkan dari latent vector melalui operasi aritmatika dalam latent space dimension [7].…”

“…Untuk menguji efek dimensi ruang laten pada citra yang dihasilkan oleh generator dibutuhkan evaluasi GAN dengan menggunakan metode yang efisien dan objektif, yaitu metode evaluasi kuantitatif [6]. Terdapat dua metrik evaluasi GAN secara kuantitaif, salah satunya adalah Fre'chet Inception Distance (FID).…”