Reproduce “NIVeL: Neural Implicit Vector Layers for Text-to-Vector Generation” 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
Vikas Thamizharasan, Difan Liu, Matthew Fisher, Nanxuan Zhao, Evangelos Kalogerakis, Michal Lukac
NIVeL, developed by researchers from UMass Amherst and Adobe Research, introduces a method for generating high-quality, editable, and resolution-independent vector graphics directly from natural language text prompts. It employs neural implicit layers and a refined Score Distillation Sampling strategy, producing outputs with cleaner geometry and improved alignment with textual descriptions compared to prior approaches.
Key insights
- Neural implicit representations provide a stable and resolution-independent method for modeling complex vector shapes and arbitrary topologies, addressing the limitations of discrete vector primitives.
- A carefully designed multi-stage initialization strategy is crucial for enabling stable optimization and achieving high-quality results when applying Score Distillation Sampling to neural implicit functions.
- Layer decomposition within the implicit representation enhances the interpretability and editability of generated vector graphics, making them more practical for design workflows.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.