Reproduce “LayerD: Decomposing Raster Graphic Designs into Layers” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tomoyuki Suzuki, Kang-Jun Liu, Naoto Inoue, Kota Yamaguchi
LayerD reconstructs flattened graphic designs as editable raster layers by repeatedly extracting visible frontmost content, filling revealed background, and refining colors in flat regions. On Crello, its layer sequences outperform baselines on reported metrics with fewer merge edits. In a user study, 15 of 21 participants rated LayerD highest on average; this does not establish recovery of original design files.
Key insights
- An iterative decomposition approach that directly extracts top layers and completes backgrounds can effectively handle the complexities of graphic designs, reducing error propagation.
- Incorporating domain-specific priors, such as the prevalence of flat colored regions, through palette-based refinement significantly enhances the accuracy and perceptual quality of decomposed layers.
- Using a non-generative inpainting model is crucial for background completion in a recursive decomposition process to prevent the introduction of new, unwanted objects, which would degrade subsequent layer extractions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.