Reproduce “Scene Graph Disentanglement and Composition for Generalizable Complex Image 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
Yunnan Wang, Ziqiang Li, Zequn Zhang, Wenyao Zhang, Baao Xie, Xihui Liu, Wenjun Zeng, Xin Jin
DisCo is a framework that generates complex images from scene graphs by disentangling layout and semantic information and then recomposing them during the generation process. The method produces visually coherent images with accurate object relationships and attributes, addressing challenges in spatial and non-spatial scene understanding for image synthesis.
Key insights
- Scene graphs can be effectively disentangled into separate layout and semantic components, enabling more precise control over image generation.
- Object-specific attention masks within diffusion models can prevent semantic confusion and attribute leakage, ensuring fine-grained attribute control.
- A multi-layered sampling approach allows for independent manipulation of individual objects and their attributes while maintaining overall scene coherence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.