Reproduce “Image Neural Field Diffusion Models” 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
Yinbo Chen, Oliver Wang, Richard Zhang, Eli Shechtman, Xiaolong Wang, Michael Gharbi
Researchers from UC San Diego, Google Research, and Adobe Research developed Image Neural Field Diffusion Models (INFD), a resolution-agnostic framework that integrates neural fields with latent diffusion models. This approach generates photorealistic, high-resolution images by learning a continuous latent representation, outperforming existing methods in detail preservation and scale consistency while enabling training on mixed-resolution datasets.
Key insights
- Neural fields can serve as an effective intermediate representation for diffusion models, enabling resolution-agnostic image generation.
- The Convolutional Local Image Function (CLIF) improves scale consistency and detail preservation in neural field rendering compared to traditional point-wise MLPs.
- Training a latent diffusion model on a neural field representation allows for efficient high-resolution image synthesis and direct training on mixed-resolution datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.