Explicit Layer Modeling for Video Object Insertion and Layer Decomposition

The paper introduces TriLayer, a large-scale triplet video dataset with aligned composite, background, and foreground videos. Foreground layers include object appearance and associated visual effects, providing explicit supervision for layered video representation learning. DBL-Diffusion uses a dual-branch diffusion architecture to jointly model RGB composites and RGBA foreground layers, with shared denoising and cross-branch interaction. Its two task-specific variants, DBL-Insert and DBL-Decompose, target realistic object insertion and foreground/background recovery. The abstract reports improved insertion fidelity and decomposition quality, but provides no quantitative results here.