ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing

TL;DR

This paper introduces ST-GAN, a novel GAN architecture using Spatial Transformer Networks for realistic geometric corrections in image compositing, enabling high-resolution image editing and object placement.

Contribution

The paper proposes a new ST-GAN architecture with an iterative warping scheme and sequential training, improving geometric correction in image compositing tasks.

Findings

Effective geometric corrections for image compositing

Applicable to high-resolution images via warp parameter transfer

Demonstrated in furniture visualization and accessory hallucination

Abstract

We address the problem of finding realistic geometric corrections to a foreground object such that it appears natural when composited into a background image. To achieve this, we propose a novel Generative Adversarial Network (GAN) architecture that utilizes Spatial Transformer Networks (STNs) as the generator, which we call Spatial Transformer GANs (ST-GANs). ST-GANs seek image realism by operating in the geometric warp parameter space. In particular, we exploit an iterative STN warping scheme and propose a sequential training strategy that achieves better results compared to naive training of a single generator. One of the key advantages of ST-GAN is its applicability to high-resolution images indirectly since the predicted warp parameters are transferable between reference frames. We demonstrate our approach in two applications: (1) visualizing how indoor furniture (e.g. from product images) might be perceived in a room, (2) hallucinating how accessories like glasses would look when matched with real portraits.