DESOBAv2: Towards Large-scale Real-world Dataset for Shadow Generation

TL;DR

This paper introduces DESOBAv2, a large-scale real-world dataset for shadow generation in image composition, enabling more realistic insertion of foreground objects by providing shadow data.

Contribution

The creation of DESOBAv2, a large-scale dataset with real outdoor images and shadow annotations, using object-shadow detection and inpainting techniques.

Findings

Enables training of shadow generation models with real-world data

Improves realism of composite images with plausible shadows

Provides a publicly available dataset for future research

Abstract

Image composition refers to inserting a foreground object into a background image to obtain a composite image. In this work, we focus on generating plausible shadow for the inserted foreground object to make the composite image more realistic. To supplement the existing small-scale dataset DESOBA, we create a large-scale dataset called DESOBAv2 by using object-shadow detection and inpainting techniques. Specifically, we collect a large number of outdoor scene images with object-shadow pairs. Then, we use pretrained inpainting model to inpaint the shadow region, resulting in the deshadowed images. Based on real images and deshadowed images, we can construct pairs of synthetic composite images and ground-truth target images. Dataset is available at https://github.com/bcmi/Object-Shadow-Generation-Dataset-DESOBAv2.