Semantic-Aware Face Deblurring With Pixel-Wise Projection Discriminator

Abstract

Most recent face deblurring methods have leveraged the distribution modeling ability of generative adversarial networks (GANs) to impose a constraint that the deblurred image should follow the distribution of sharp ground-truth images. However, generating sharp face images with high fidelity and realistic properties from a blurry face image remains challenging under the GAN framework. To this end, we focus on modeling the joint distribution of sharp face images and segmentation label maps for face image deblurring in a GAN framework. We propose a semantic-aware pixel-wise projection (SAPP) discriminator that models pixel-label matching with semantic label map information and generates a pixel-wise probability map of realness for the input image as well as a per-image probability. Moreover, we introduce a predictionweighted (PW) loss to focus on erroneous pixels in the output of the decoder, using per-pixel real/fake probability map to re-weight the contribution of each pixel in the decoder. Furthermore, we present a coarseto-fine training technique for the generator, which encourages the generator to focus on global consistency in the early training stages and local details in the later stages. Extensive experimental results show that our method outperforms existing methods both quantitatively and qualitatively in terms of perceptual image quality.INDEX TERMS Face image deblurring, semantic-aware pixel-wise projection discriminator, predictionweighted loss.

GammaGAN: Gamma-Scaled Class Embeddings for Conditional Video Generation

“…This is typically done by directly injecting it into the input [13] or by injecting it into a hidden layer [14][15][16][17][18]. However, projection-based methods [12,[19][20][21] have been developed to effectively utilize class information for cGANs. This method involves generating two types of embeddings (feature vectors and class embeddings) and then calculating the inner products between them to measure their similarities.…”

“…Techniques for utilizing class information in discriminators can generally be categorized into two types: injecting a class label directly and employing auxiliary classifiers. The former is commonly achieved by concatenating [11,[13][14][15][16][17][18] or by projecting the class information using class embeddings [12,[19][20][21]. On the other hand, auxiliary classifiers [22][23][24][25][26] aim to use class information effectively through classification.…”