A Likely Geometry of Generative Models -- Application to ControlNet

Experimential results on ControlNet for method described in https://github.com/FrederikMR/likely_geometry. Please se that GitHub for details.

Installation and Requirements

The implementations in the GitHub is Python 3.10.12 and has been implemented in both JAX and PyTorch.

Reproducing Experiments

All experiments can be re-produced by running the notebooks and the run_interpolation.py for the given data, method and hyper-parameters.

Reference

If you want to use the algorithm or the method proposed in the paper for scientific purposes, please cite:

@misc{rygaard2026likelygeometrygenerativemodels,
      title={A Likely Geometry of Generative Models}, 
      author={Frederik Möbius Rygaard and Shen Zhu and Yinzhu Jin and Søren Hauberg and Tom Fletcher},
      year={2026},
      eprint={2510.26266},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2510.26266}, 
}