Reproduce “Discovering Latent Graphs with GFlowNets for Diverse Conditional 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

Bailey Trang, Parham Saremi, Alan Q. Wang, Fangrui Huang, Zahra TehraniNasab, Amar Kumar, Tal Arbel, Li Fei-Fei and 1 more

The Rainbow framework enables diverse conditional image generation by integrating latent graphs and Generative Flow Networks (GFlowNets) to decompose input conditions into multiple distinct latent representations. This approach generates a wider range of plausible images for both natural and medical data, demonstrated by improved diversity and quality metrics and enhanced performance in downstream tasks.

Key insights

  • Explicitly decomposing ambiguous input conditions into structured, diverse latent representations using GFlowNets effectively addresses multimodal uncertainty in conditional image generation.
  • Generative Flow Networks (GFlowNets) are a suitable mechanism for exploring the high-quality, diverse solution space of uncertain conditions by sampling trajectories proportional to their reward.
  • The learned latent graphs acquire interpretable semantic meanings, allowing specific edges to control the presence or alteration of distinct visual features in generated images.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.