IR-GAN: Image Manipulation with Linguistic Instruction by Increment Reasoning

TL;DR

IR-GAN is a novel model that enhances image manipulation by reasoning the consistency between visual and semantic increments, improving the logical coherence of generated images based on linguistic instructions.

Contribution

The paper introduces IR-GAN, which incorporates semantic and visual increment reasoning with a new discriminator to improve multimodal conditional image generation.

Findings

IR-GAN effectively measures and enforces consistency between image and instruction increments.

Experimental results demonstrate IR-GAN's superior performance on two datasets.

Visualization confirms improved logical coherence in generated images.

Abstract

Conditional image generation is an active research topic including text2image and image translation. Recently image manipulation with linguistic instruction brings new challenges of multimodal conditional generation. However, traditional conditional image generation models mainly focus on generating high-quality and visually realistic images, and lack resolving the partial consistency between image and instruction. To address this issue, we propose an Increment Reasoning Generative Adversarial Network (IR-GAN), which aims to reason the consistency between visual increment in images and semantic increment in instructions. First, we introduce the word-level and instruction-level instruction encoders to learn user's intention from history-correlated instructions as semantic increment. Second, we embed the representation of semantic increment into that of source image for generating target image, where source image plays the role of referring auxiliary. Finally, we propose a reasoning discriminator to measure the consistency between visual increment and semantic increment, which purifies user's intention and guarantees the good logic of generated target image. Extensive experiments and visualization conducted on two datasets show the effectiveness of IR-GAN.