Reproduce “Encoding Semantic Priors into the Weights of Implicit Neural Representation” 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

Zhicheng Cai, Qiu Shen

A reparameterization framework called SPW enables Implicit Neural Representations (INRs) to integrate semantic information directly into their weights. This method improves INR performance across image fitting, medical image reconstruction, and 5D novel view synthesis tasks without adding computational overhead during inference.

Key insights

  • Explicitly encoding semantic priors into INR weights reduces weight redundancy and increases weight entropy, leading to more diverse internal representations.
  • The effectiveness of the approach stems from integrating both high-level and low-level semantic features from a pre-trained CNN.
  • The reparameterization process allows for enhancing INR capabilities without incurring additional runtime costs, making it practically efficient.

Summary by alphaXiv.

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