Reproduce “Illustrator's Depth: Monocular Layer Index Prediction for Image Decomposition” 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
Nissim Maruani, Peiying Zhang, Siddhartha Chaudhuri, Matthew Fisher, Nanxuan Zhao, Vladimir G. Kim, Pierre Alliez, Mathieu Desbrun and 1 more
Researchers introduce "Illustrator's Depth," a novel method for decomposing flat images into discrete, ordered layers that reflect an artist's compositional intent rather than physical depth. The approach significantly improves layered vectorization, enabling the creation of editable vector graphics from text prompts and facilitating automatic 3D relief generation.
Key insights
- Compositional depth, optimized for editability, is a more suitable and interpretable representation for artistic image decomposition than physical depth, enabling superior layering results.
- Initializing the model with a physical depth prior (e.g., Depth Pro weights) and training on carefully curated, consistent layer data significantly improves the network's ability to infer compositional order.
- Directly optimizing for discrete layer indices (1 to N) rather than disparity leads to better depth transitions and more balanced optimization across all layers in an illustration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.