![]()
Face morph age progression applications generator#The original inputs are decomposed into multi-level wavelet coefficients before being fed to the generator with multi-level encoders. In our approach, we design a novel multi-level generator combined with a wavelet packet transform (WPT) module for facial features extraction. However, this approach can only provide facial attributes at no more than two levels. GLCA-GAN (Li et al., 2018) leveraged one global and three local generators to capture facial features at different scales for face aging. Intuitively, feeding multi-level features of original inputs to multi-level generator should be able to directly improve the performance of the model in face aging but few works dig in this way. Wavelet-GAN (Liu et al., 2019) also shows that a Wavelet-based decomposition of inputs in advance can help the multi-level discriminator to capture age-related texture at multiple scales. Recently, the work related to multi-level GAN (Yang et al., 2018, Liu et al., 2019) shows that facial features extracted from images at multiple scales by discriminators can force the generator to synthesize vivid aging effects by back-propagation of the adversarial/condition loss from the discriminators to the generator. However, in these works, a well-trained StyleGAN is needed. Besides, some recent works (Alaluf et al., 2021, Hou et al., 2022) have proposed to consider face aging as a style transformation and designed a Conditional GAN to learn the translation of age in the latent space of StyleGAN. The current Cycle-GAN (Zhu et al., 2017) based methods (Song et al., 2018, Sun et al., 2020) can combine both face progression and regression with a single model, but suffer from blurring results, less fine-grained details and even artifacts due to the challenge of predicting faces falls in different age groups with one model. However, empirical experiments show that adding a pretrained deep neural network (e.g. a VGG-Face descriptor (Parkhi et al., 2015)) will increase the training time dramatically. Additionally, most of the current methods (Wang et al., 2018, Li et al., 2018, Yang et al., 2018, Liu et al., 2019, Fang et al., 2020, Li et al., 2020a) adopt a pretrained neural network for identity preserving. However, these methods reduced the efficiency of the aging models. Many previous works (Wang et al., 2018, Yang et al., 2018, Liu et al., 2019, Huang et al., 2020) trained different GANs for age translation with different source or target age group. In the area of face aging, most generative models are based on Conditional GAN. Especially, the recent advent of Generative Adversarial Networks (GANs), which have obtained an amazing achievement in generating photo-realistic images (Goodfellow et al., 2014, Arjovsky et al., 2017, Huang et al., 2017, Radford et al., 2015, Zhao et al., 2016), has opened a new door to the face manipulating technologies.Īlthough the recent proposed StyleGAN (Karras et al., 2019, Karras et al., 2020) have achieved superior results for image synthesis with amazing photo-realism, manipulation of attribute specified in these uncontrolled GANs is challenging. Although difficult it is, face aging has achieved great progress owing to the rapid development of deep neural networks. ![]() However, face aging is an intractable task owing to the lack of image set of the same person over a long age span as well as the variants of face poses, the change of illumination, and the existence of occlusion (Liu et al., 2017). Face aging originated from the need of finding missing children has shown significance for cross-age recognition and recreation applications. ![]() Face morph age progression applications verification#Our model can outperform most of the existing approaches include the state-of-the-art techniques on two benchmark aging databases in terms of both aging accuracy and identity verification confidence, demonstrating the effectiveness and superiority of our method.įace aging including progression and regression has attracted much attention from the community of computer vision in last decade. Experiment results on both MORPH and CACD show that the application of multi-level generator can improve the identity preserving effects in face aging and reduce the training time significantly by eliminating the rely of an identity preserving module. The proposed model consists of one generator with three-level encoders and three levels of discriminators with an age and a gender classifier on top of each discriminator. In this work, we propose a cycle-consistent loss based method for face aging with wavelet-based multi-level facial attributes extraction from both generator and discriminators. Previous works usually rely on an extra pre-trained module for identity preserving and multi-level discriminators for fine-grained features extraction. Age accuracy and identity preserving are two important indicators for face aging. Face aging has received increasing attention from the computer vision community due to wide applications in the real world. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |