Masters Thesis Defense: Jun Shao

Face aging has received increasing attention from the computer vision community due to wide applications in the real world. Age accuracy and identity preserving are two important indicators for face aging. Previous works usually rely on an extra pretrained module for identity preserving and multi-level discriminators for fine-grained features extraction. 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. 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. 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. 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. 

Expression translation has received increasing attention from the computer vision community due to its wide applications in the real world. However, expression synthesis is hard because of the non-linear properties of facial skin and muscle caused by different expressions. A recent study showed that the practice of using the same generator for both forward prediction and backward reconstruction as in current conditional GANs would force the generator to leave a potential ”noise” in the generated images, therefore hindering the use of the images for further tasks. To eliminate the interference and break the unwanted link between the first and second translation, we design a parallel training mechanism with two generators that perform the same first translation but work as a reconstruction model for each other. Additionally, inspired by the successful application of wavelet-based multi-level Generative iii Adversarial Networks(GANs) in face aging and progressive training in geometric conversion, we further design a novel wavelet-based multi-level Generative Adversarial Network (WP2-GAN) for expression translation with a large gap based on a progressive and parallel training strategy. Extensive experiments show the effectiveness of our approach for expression translation compared with the state-of-the-art models by synthesizing photo-realistic images with high fidelity and vivid expression effect.