Brief

Researchers have developed a new framework for decomposing real-world images into layers, addressing limitations in current generative models that are primarily effective in graphic design. Their approach includes an Agent-driven Data Decomposition (ADD) pipeline to create a large dataset of over 100,000 layered images, named LiWi-100k. The proposed model enhances photometric fidelity and alpha boundary accuracy by explicitly modeling illumination effects and using a degradation-restoration objective for boundary correction. Experiments show this method achieves state-of-the-art performance in natural image decomposition. AI