Metraipolis › archive › dc_f85372c879ed

Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition — arXiv:2610.41789

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.41789
License (as archived)
CC0-1.0
Archived
2026-11-02T05:03:48.362Z
SHA-256 of the archived text
cebe7186ab90f1e6c7eee3e0fa2d630bc936ad878600097d62dee35edf301a29
Stable URL of this page
https://metraipolis.com/documento/dc_f85372c879ed

How to cite

Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition — arXiv:2610.41789 — arXiv (en), https://arxiv.org/abs/2610.41789 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f85372c879ed (Metraipolis archive dc_f85372c879ed, sha256 cebe7186ab90f1e6c7eee3e0fa2d630bc936ad878600097d62dee35edf301a29)
BibTeX
@misc{dcf85372c879ed,
  title = {Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition — arXiv:2610.41789},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.41789},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f85372c879ed; sha256 cebe7186ab90f1e6c7eee3e0fa2d630bc936ad878600097d62dee35edf301a29},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.41789. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multi-head attention in Vision Transformers (ViTs) allows independent attention patterns across heads, but without explicit regularization, heads may learn redundant patterns rather than complementary roles. Correlation-based penalties encourage head disentanglement, yet apply uniform regularization across all training samples. In face recognition (FR), hard samples whose embeddings lie close to competing identities are critical for optimization and generalization. Motivated by this, we propose ViT-HarDis, which conditions attention head disentanglement on sample difficulty: samples with high cosine similarity to negative class centers receive stronger head decorrelation, enforcing specialization where identity ambiguity is highest. Experiments show improvements over both the baseline and uniform disentanglement on most benchmarks. Our analysis shows that ViT-HarDis reduces the similarity of hard samples to competing class centers while decreasing inter-head attention similarity, and that applying disentanglement only to the final transformer layers outperforms regularizing the entire network. Trained on WebFace4M, ViT-HarDis achieves SOTA results among ViT-based methods on IJB-B, IJB-C, and TinyFace Rank-1, while achieving the best IJB-B and TinyFace results when trained on MS1MV3, improving FR without architectural modifications or additional inference cost. Code: this https URL .

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/2610.41789 — 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 2026-11-02T04:49:13.273Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).