Abstract:Generative Adversarial Networks (GANs) have recently achieved significant improvement on paired/unpaired image-to-image translation, such as photo$\rightarrow$ sketch and artist painting style transfer. However, existing models can only be capable of transferring the low-level information (e.g. color or texture changes), but fail to edit high-level semantic meanings (e.g., geometric structure or content) of objects. On the other hand, while some researches can synthesize compelling real-world images given a class label or caption, they cannot condition on arbitrary shapes or structures, which largely limits their application scenarios and interpretive capability of model results. In this work, we focus on a more challenging semantic manipulation task, which aims to modify the semantic meaning of an object while preserving its own characteristics (e.g. viewpoints and shapes), such as cow$\rightarrow$sheep, motor$\rightarrow$ bicycle, cat$\rightarrow$dog. To tackle such large semantic changes, we introduce a contrasting GAN (contrast-GAN) with a novel adversarial contrasting objective. Instead of directly making the synthesized samples close to target data as previous GANs did, our adversarial contrasting objective optimizes over the distance comparisons between samples, that is, enforcing the manipulated data be semantically closer to the real data with target category than the input data. Equipped with the new contrasting objective, a novel mask-conditional contrast-GAN architecture is proposed to enable disentangle image background with object semantic changes. Experiments on several semantic manipulation tasks on ImageNet and MSCOCO dataset show considerable performance gain by our contrast-GAN over other conditional GANs. Quantitative results further demonstrate the superiority of our model on generating manipulated results with high visual fidelity and reasonable object semantics.

Semantics-Enhanced Adversarial Nets for Text-to-Image Synthesis.

SIMGAN: Photo-Realistic Semantic Image Manipulation Using Generative Adversarial Networks.

Fine-grained Semantic Constraint in Image Synthesis

Semantics Disentangling for Text-to-Image Generation

Multi-Semantic Fusion Generative Adversarial Network for Text-to-Image Generation

Dual Attention GANs for Semantic Image Synthesis

Linear Semantics in Generative Adversarial Networks

DSE-GAN: Dynamic Semantic Evolution Generative Adversarial Network for Text-to-Image Generation

Text-to-image Generation Based on Spatial-Channel Attention and Semantic Redescription

Multi-Sentence Auxiliary Adversarial Networks for Fine-Grained Text-to-Image Synthesis

SAM-GAN: Self-Attention supporting Multi-stage Generative Adversarial Networks for text-to-image synthesis

Semantic Image Synthesis via Adversarial Learning

Perceptual Pyramid Adversarial Networks for Text-to-Image Synthesis.

Edge Guided GANs with Contrastive Learning for Semantic Image Synthesis

Generative Semantic Manipulation with Contrasting GAN

Object-driven Text-to-Image Synthesis via Adversarial Training

Semantic Draw Engineering for Text-to-Image Creation

Disentangling for Text-to-Image Generation

Art design integrating visual relation and affective semantics based on Convolutional Block Attention Mechanism-generative adversarial network model

Local and Global GANs with Semantic-Aware Upsampling for Image Generation

Semantic Image Analogy with a Conditional Single-Image GAN