GauGAN: semantic image synthesis with spatially adaptive normalization

By Taesung Park, Ming-Yu Liu, Ting-Chun Wang, Jun-Yan Zhu. 2019

154 citations recorded by the source corpus.

We propose GauGAN, a GAN-based image synthesis model that can generate photo-realistic images given an input semantic layout. It is built on spatially-adaptive normalization, a simple but effective normalization layer. Previous methods directly feed the semantic layout as input to the deep network, which is then processed through stacks of convolution, normalization, and non-linearity layers. We show that this is sub-optimal as the normalization layers tend to "wash away" semantic information. To address the issue, we propose using the input layout for modulating the activations in normalization layers through a spatially-adaptive, learned transformation. Our proposed method outperforms the previous methods by a large margin. Furthermore, the new method enables natural extension to control the style of the synthesized images. Given a style guide image, our style encoder network captures the style into a latent code, which our image generator network combines with the semantic layout via spatially-adaptive normalization to generate a photo-realistic image that respects both the style of the guide image and content of the semantic layout. Our method will enable people without drawing skills to effectively express their imagination. GauGAN in the inference time is a simple convolutional neural network. It runs real-time on most modern GPU cards. GauGAN is one of the recent research efforts in advancing GANs for real-time image rendering. We believe this is of interest to the SIGGRAPH and real-time communities.

Subfields: Computer Graphics and Computer-Aided Design, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Graphics and Computer-Aided Design

In Computer Vision and Pattern Recognition

Citations in the corpus

Mimicry: Towards the Reproducibility of GAN Research (2020, background): “In particular, GANs are known for their creative applications in areas such as image synthesis [36, 17, 32, 18] image-to-image translation [42, 43, 23, 24], video-to-video synthesis [39, 38, 7], and even 3D view synthesis [29, 30, 40].”
Creative Sketch Generation (2020, background): “Most existing datasets contain sketches drawn by humans to realistically mimic common objects (Eitz et al., 2012; Sangkloy et al., 2016; Jongejan et al., 2016; Wang et al., 2019).”
Exploring Object-Centric Temporal Modeling for Efficient Multi-View 3D Object Detection (2023, methodology): “Motivated by the taskspecific control in generative model [9, 40, 55], we adopt a conditional layer normalization to model the movement of the objects.”
Layout2image: Image Generation from Layout (2020, methodology): “GauGAN (Park et al. 2019) is proposed to synthesize high-resolution images with realistic details.”
Sketchformer: Transformer-Based Representation for Sketched Structure (2020, background): “For example, fusion with additional modalities might enable sketch driven photo generation [41] using complex sketches, or with a language embedding for novel sketch synthesis applications.”