Tezislər / Theses



Yüklə 17,55 Mb.
Pdf görüntüsü
səhifə153/493
tarix02.10.2023
ölçüsü17,55 Mb.
#151572
1   ...   149   150   151   152   153   154   155   156   ...   493
BHOS Tezisler 2022 17x24sm

THE 3
rd
 INTERNATIONAL SCIENTIFIC CONFERENCES OF STUDENTS AND YOUNG RESEARCHERS 
dedicated to the 99
th
anniversary of the National Leader of Azerbaijan Heydar Aliyev
161
Feature mapping methodology to transfer from so-called “parent” 
domain into “child” feature domain is motivated from DNA-Net architecture 
which combines Conditional Adversarial Autoencoders and genetic process 
simulations to realize child facial image generation [5]. 
As the method is modelling the relationship between abstract 
convolutional features of parent facial images and child latent features, it can 
successfully predict the decoupled representation of child image and then 
later feed it as style vector to the StyleGAN generator. StyleGAN generator 
will take those spatially invariant style vectors and control adaptive instance 
normalization blocks which are normalizing outputs of convolutional layers 
corrupted by single-channel uncorrelated Gaussian noise. Child gender 
information will be processed during feature mapping stage and will be 
utilized to apply learned weighting to aggregate transformed features of 
parent images. As abstract child facial features are obtained, it will be 
relatively easy to manipulate those vectors and generate facial images of 
needed age.
Figure 10.
 Visualization of Latent Feature Mapper Network 


THE 3
rd
 INTERNATIONAL SCIENTIFIC CONFERENCES OF STUDENTS AND YOUNG RESEARCHERS 
dedicated to the 99
th
anniversary of the National Leader of Azerbaijan Heydar Aliyev
162
References
[1] - NVIDIA, A Style-Based Generator Architecture for Generative Adversarial Networks, 
[2] Authors: Tero Karras, Samuli Laine, Timo Aila 
[3] - Tel-Aviv University, Designing an Encoder for StyleGAN Image Manipulation, Authors: 
[4] Omer Tov, Yuval Alaluf, Yotam Nitzan, Or Patashnik, Daniel Cohen-Or 
[5] - Penta AI, Tel-Aviv University, Encoding in Style: a StyleGAN Encoder for Image-to- 
Image Translation, Authors: Elad Richardson, Yuval Alaluf, Or Patashnik, Yotam 
Nitzan, Yaniv Azar, Stav Shapiro, Daniel Cohen-Or 
[6] - IEEE, Heredity-Aware Child Face Image Generation with Latent Space Disen-
tanglement, Authors: Xiao Cui, Wengang Zhou, Yang Hu, Weilun Wang and 
Houqiang Li 
[7] – What will your child look like? DNA-Net: Age & Gender Aware Kin Face 
Synthesizer, Authors: Pengyu Gao, Siyu Xia, Joseph Robinson, Junkang Zhang, 
Chao Xia, Ming Shao, YUN FU 

Yüklə 17,55 Mb.

Dostları ilə paylaş:
1   ...   149   150   151   152   153   154   155   156   ...   493




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©azkurs.org 2024
rəhbərliyinə müraciət

gir | qeydiyyatdan keç
    Ana səhifə


yükləyin