Shadow Generation with Decomposed Mask Prediction and Attentive Shadow Filling

TL;DR

This paper introduces DMASNet, a two-stage network for generating realistic shadows in image composition, utilizing a large dataset and decomposed mask prediction to improve visual quality and generalization.

Contribution

The work presents a novel two-stage network with decomposed mask prediction and attentive shadow filling, along with a new large-scale dataset for shadow generation.

Findings

DMASNet outperforms existing methods in visual quality.

The model generalizes well to real composite images.

The large dataset enhances training effectiveness.

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 shadows for the inserted foreground object to make the composite image more realistic. To supplement the existing small-scale dataset, we create a large-scale dataset called RdSOBA with rendering techniques. Moreover, we design a two-stage network named DMASNet with decomposed mask prediction and attentive shadow filling. Specifically, in the first stage, we decompose shadow mask prediction into box prediction and shape prediction. In the second stage, we attend to reference background shadow pixels to fill the foreground shadow. Abundant experiments prove that our DMASNet achieves better visual effects and generalizes well to real composite images.