Reproduce “LR-GAN: Layered Recursive Generative Adversarial Networks for 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
Jianwei Yang, Anitha Kannan, Dhruv Batra, Devi Parikh
Researchers at Virginia Tech and Facebook AI Research developed LR-GAN, a generative adversarial network that improves image realism by explicitly modeling scene structure as layered backgrounds and foreground objects. The model recursively generates object appearance, shape, and pose, leading to more distinct and recognizable foreground elements within generated images.
Key insights
- Explicitly modeling the layered composition of natural scenes, by separately generating backgrounds and foregrounds, enhances the realism and recognizability of objects in generated images.
- Factorizing object properties into appearance, shape, and pose enables more flexible and realistic object representations, allowing the model to apply transformations and learn object masks without explicit supervision.
- A recursive generation process that considers previously generated elements helps maintain contextual relevance and spatial coherence for new foreground objects within a scene.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.