结合Pixel2style2Pixel的年龄转化方法

doi:10.3778/j.issn.1002-8331.2304-0007

摘要/Abstract

摘要： 年龄转化在刑侦、人脸识别等领域有着重要作用。常见的年龄转化方法需要使用成对带有年龄注释的数据集进行训练，并且存在生成图像质量低、年龄语义信息不够解耦等问题。针对上述问题，在Pix2style2Pix的训练框架中，引入年龄识别、contextual损失函数，对整体损失函数做出符合年龄转化的改进，提取年龄信息并保证图像质量。改进编码网络配合损失函数对潜在空间的图像进行编辑，提出一种基于Pixel2style2Pixel的年龄转化方法。通过FFHQ、CelebA数据集，对所提方法进行验证，实验结果表明，在不采用成对年龄注释的训练集下，改进后的损失函数能生成更符合期望年龄的转化图像，人脸相似度距离为0.346、FID为45.69、SSIM为0.593?6、PSNR为19.64?dB，均优于对比方法，证明所提方法能够生成高质量、年龄语义高度解耦的转化结果。

关键词: Pixel2style2Pixel, 人脸年龄转化, StyleGAN, 损失函数, 图像处理

Abstract: Age transformation has an important role in criminal investigation, face recognition and other fields. Common age transformation methods need to use pairs of datasets with age annotations for training, and have problems such as low quality of generated images and insufficient decoupling of age semantic information. To address the above problems, in the training framework of Pix2style2Pix, age recognition, contextual loss function is introduced to make improvements to the overall loss function in line with age transformation to extract age information and ensure image quality. Secondly, it improves the coding network with loss function to edit the image in potentials space, and proposes an age transformation method based on Pixel2style2Pixel. The proposed method is validated with FFHQ and CelebA datasets, and the experimental results show that the improved loss function can generate transformed images that better match the desired age with face similarity distance of 0.346, FID of 45.69, SSIM of 0.593?6, and PSNR of 19.64?dB, all of which are better than the comparison method under the training set without pairwise age annotation The proposed method is proved to be able to generate high-quality, highly decoupled age-semantic transformation results.

Key words: Pixel2style2Pixel, face aging, StyleGAN, loss function, image processing

桂列林, 黄山, 印月. 结合Pixel2style2Pixel的年龄转化方法[J]. 计算机工程与应用, 2024, 60(14): 162-174.

GUI Lielin, HUANG Shan, YIN Yue. Age Transformation Method Combined with Pixel2style2Pixel[J]. Computer Engineering and Applications, 2024, 60(14): 162-174.

参考文献

[1] DUONG N C, LUU K, QUACH G K, et al. Longitudinal face modeling via temporal deep restricted Boltzmann machines[C]//Proceedings of the 2016 IEEE International Conference on Computer Vision, 2016: 5772-5780.
[2] WANG W, CUI Z, YAN Y, et al. Recurrent face aging[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, 2016: 2378-2386.
[3] WANG W, YAN Y, CUI Z, et al. Recurrent face aging with hierarchical autoregressive memory[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 41(3): 654-668.
[4] HELJAKKA A, SOLIN A, KANNALA J. Recursive chaining of reversible image-to-image translators for face aging[C]//Proceedings of the International Conference on Advanced Concepts for Intelligent Vision Systems, 2018: 309-320.
[5] LI P, HUANG H, HU Y, et al. UVA: a universal variational framework for continuous age analysis[J]. arXiv:1904.00158, 2019.
[6] YANG H, HUANG D, WANG Y, et al. Learning face age progression: a pyramid architecture of GANs[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, 2018: 31-39.
[7] LIU S, SUN Y, ZHU D, et al. Face aging with contextual generative adversarial nets[C]//Proceedings of the 25th ACM International Conference on Multimedia, 2017: 82-90.
[8] PALSSON S, AGUSTSSON E, TIMOFTE R, et al. Generative adversarial style transfer networks for face aging[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2018: 2084-2092.
[9] ANTIPOV G, BACCOUCHE M, DUGELAY J L. Face aging with conditional generative adversarial networks[C]//Proceedings of the 2017 IEEE International Conference on Image Processing, 2017: 2089-2093.
[10] ZHANG Z, SONG Y, QI H.Age progression/regression by conditional adversarial autoencoder[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition, 2017: 5810-5818.
[11] HE Z, KAN M, SHAN S, et al. S2GAN: share aging factors across ages and share aging trends among individuals[C]//Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition, 2019: 9440-9449.
[12] OR-EL R, SENGUPTA S, FRIED O, et al. Lifespan age transformation synthesis[C]//Proceedings of the 16th European Conference on Computer Vision, 2020: 739-755.
[13] WANG Z, TANG X, LUO W, et al. Face aging with identity-preserved conditional generative adversarial networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7939-7947.
[14] WU Y, WANG R, GONG M, et al. Adversarial UV-transformation texture estimation for 3D face aging[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 32(7): 4338-4350.
[15] DUAN M, LI K, LIAO Q, et al. DEF-Net: a face aging model by using different emotional learnings[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 32(5): 3012-3022.
[16] 于海波. 基于生成对抗网络的人脸老化算法研究[D]. 北京: 北京交通大学, 2021.
YU H B. Research on face aging algorithm based on generative adversarial network[D]. Beijing: Beijing Jiaotong University, 2021.
[17] LI Q, LIU Y, SUN Z.Age progression and regression with spatial attention modules[C]//Proceedings of the National Conference on Artificial Intelligence, 2020: 11378-11385.
[18]ABDAL R, ZHU P, MITRA N J, et al.Styleflow: attribute-conditioned exploration of StyleGAN-generated images using conditional continuous normalizing flows[J]. ACM Transactions on Graphics, 2021, 40(3): 1-21.
[19] H?RK?NEN E, HERTZMANN A, LEHTINEN J, et al.GANspace: discovering interpretable GAN controls[C]//Advances in Neural Information Processing Systems, 2020, 33: 9841-9850.
[20] LIU Y, LI Q, SUN Z, et al. Style intervention: how to achieve spatial disentanglement with style-based generators?[J]. arXiv:2011.09699, 2020.
[21] SHEN Y, GU J, TANG X, et al.Interpreting the latent space of GANs for semantic face editing[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 9243-9252.
[22] PANTRAKI E, KOTROPOULOS C. Face aging using global and pyramid generative adversarial networks[J]. Machine Vision and Applications, 2021, 32(4): 82.
[23] HUANG Z, CHEN S, ZHANG J, et al. AgeFlow: conditional age progression and regression with normalizing flows[J].arXiv:2105.07239, 2021.
[24] KARRAS T, LAINE S, AILA T. A style-based generator architecture for generative adversarial networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 4401-4410.
[25] HE S, LIAO W, YANG M Y, et al. Disentangled lifespan face synthesis[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 3877-3886.
[26] KARRAS T, LAINE S, AITTALA M, et al. Analyzing and improving the image quality of StyleGAN[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 8110-8119.
[27] ABDAL R, QIN Y, WONKA P. Image2StyleGAN: how to embed images into the StyleGAN latent space?[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 4432-4441.
[28] RICHARDSON E, ALALUF Y, PATASHNIK O, et al. Encoding in style: a StyleGAN encoder for image-to-image translation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 2287-2296.
[29] WU Z, LISCHINSKI D, SHECHTMAN E. StyleSpace analysis: disentangled controls for StyleGAN image generation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 12863-12872.
[30] RAMANATHAN N, CHELLAPPA R.Modeling age progression in young faces[C]//Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2006: 387-394.
[31] GUO C, FAN B, ZHANG Q, et al.AugFPN: improving multi-scale feature learning for object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 12595-12604.
[32] ALALUF Y, PATASHNIK O, COHEN-OR D. Only a matter of style: age transformation using a style-based regression model[J]. ACM Transactions on Graphics, 2021, 40(4): 1-12.
[33] MECHREZ R, TALMI I, ZELNIK-MANOR L. The contextual loss for image transformation with non-aligned data[C]//Proceedings of the European Conference on Computer Vision, 2018: 768-783.
[34] DENG J, GUO J, XUE N, et al. ArcFace: additive angular margin loss for deep face recognition[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 4690-4699.
[35] ZHANG R, ISOLA P, EFROS A A, et al. The unreasonable effectiveness of deep features as a perceptual metric[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 586-595.
[36] LIU Z, LUO P, WANG X, et al. Deep learning face attributes in the wild[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015: 3730-3738.
[37] WRIGHT L, DEMEURE N. Ranger21: a synergistic deep learning optimizer[J].arXiv:2106.13731, 2021.
[38] ZHANG M, LUCAS J, BA J, et al. Lookahead optimizer: k steps forward, 1 step back[J]. arXiv:1907.08610, 2019.
[39] PARKHI O M, VEDALDI A, ZISSERMAN A .Deep face recognition[C]//Proceedings of the British Machine Vision Conference, 2015.
[40] HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANs trained by a two time-scale update rule converge to a local nash equilibrium[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 6629-6640.