Weight Space Representation Learning on Diverse NeRF Architectures (ICLR 2026)

This repository contains the models for the paper Weight Space Representation Learning on Diverse NeRF Architectures, accepted at ICLR 2026. The paper proposes a framework that is capable of processing NeRFs with diverse architectures (MLPs, tri-planes, and hash tables) by training a graph metanetwork to obtain an architecture-agnostic latent space.

Repo content

• l_con/best.pt: \(\mathcal{L}_\text{C}\) model
• l_rec/best.pt: \(\mathcal{L}_\text{R}\) model
• l_rec_con/best.pt: \(\mathcal{L}_\text{R+C}\) model
• llana: LLaNA model trained on \(\mathcal{L}_\text{R+C}\) embeddings

Cite us

If you find our work useful, please cite us:

@inproceedings{ballerini2026weight,
  title = {Weight Space Representation Learning on Diverse {NeRF} Architectures},
  author = {Ballerini, Francesco and Zama Ramirez, Pierluigi and Di Stefano, Luigi and Salti, Samuele},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year = {2026}
}