Shadow Generation with Decomposed Mask Prediction and Attentive Shadow Filling

• DOI: 10.1609/aaai.v38i6.28326
• Publication Date: 2024-03-25T09:47:46Z
• Authors: Xinhao Tao; Junyan Cao; Yan Hong; Li Niu
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

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