Laura C.
Diaz-Delgado

Frozen CLIP Priors for Robust Self-Supervised Poisson Inverse Problems

Laura C. Diaz-Delgado, Emmanuel Martinez, Henry Arguello

Universidad Industrial de Santander · Malmö, Sweden · September 2026

Poster Session 3 · Fri 11 Sept, 10:30–12:30 CEST · ExHall, board #50

An ADMM-inspired unrolled plug-and-play solver for Poisson inverse problems that decouples a closed-form data-consistency update from a parameter-efficient prior. The prior is a lightweight decoder over frozen CLIP RN50 dense multi-scale features, and self-supervision combines GR2R measurement-domain re-corruption with an equivariant-imaging regulariser over virtual acquisitions — so no clean ground truth is ever required.

@inproceedings{diazdelgado2026frozenclip,
  title     = {Frozen {CLIP} Priors for Robust Self-Supervised
               Poisson Inverse Problems},
  author    = {Diaz-Delgado, Laura C. and Martinez, Emmanuel
               and Arguello, Henry},
  booktitle = {Proceedings of the European Conference on
               Computer Vision (ECCV)},
  year      = {2026},
  eprint    = {2608.20524},
  archivePrefix = {arXiv},
  primaryClass  = {eess.IV}
}