Metraipolis › archive › dc_e79ee384c481

Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery — arXiv:2610.07337

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.07337
License (as archived)
CC0-1.0
Archived
2026-10-07T13:07:05.886Z
SHA-256 of the archived text
b6bae60669a8ac309914a4a03b5f3706651e3c21620281a360778fc869d8dce5
Stable URL of this page
https://metraipolis.com/documento/dc_e79ee384c481

How to cite

Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery — arXiv:2610.07337 — arXiv (en), https://arxiv.org/abs/2610.07337 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e79ee384c481 (Metraipolis archive dc_e79ee384c481, sha256 b6bae60669a8ac309914a4a03b5f3706651e3c21620281a360778fc869d8dce5)
BibTeX
@misc{dce79ee384c481,
  title = {Compositional Concept Erasure in Text-to-Image Diffusion Models via Hierarchically Grounded Semantic Surgery — arXiv:2610.07337},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.07337},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e79ee384c481; sha256 b6bae60669a8ac309914a4a03b5f3706651e3c21620281a360778fc869d8dce5},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.07337. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Removing copyrighted, unsafe, or user-specified concepts from a deployed text-to-image diffusion model is now a practical requirement. Weight-editing methods can suppress fixed targets, but they require per-target retraining and modify the model checkpoint. Training-free methods, on the other hand, are deployment-friendly, but they suffer from text-side routing failures on compositional prompts. In such prompts, the erase target may be invoked through a related class rather than its lexical name, and its modifiers may migrate onto preserved objects. This paper proposes Hierarchically Grounded Semantic Surgery (HGSS), a training-free framework for compositional concept erasure. The framework lifts both the routing signal and the edit operator used by text-side erasure. First, hierarchical span grounding resolves erase-target spans through lexical, taxonomic, and semantic evidence, while guarding against broad-hypernym and compound-head false positives. Second, dynamic attribute binding refines the text conditioning during early denoising via a counterfactual reference and a preserve-aware cross-attention objective, keeping surviving attribute-noun bindings intact. HGSS selectively removes the erase target without updating model weights or adding learned parameters. On SEE, HGSS cuts hierarchical evasion from 29.54 to 10.02 and roughly halves pairwise attribute leakage, achieving the best Neighbor E and AttrP scores among the reported erasure methods. On UnlearnCanvas, HGSS slightly improves the six-metric average over the matched Semantic Surgery baseline, reaching state-of-the-art.

Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.07337 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T12:37:32.877Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).