Metraipolis › archive › dc_f04a67f25d43

ViTok: Improving Dense Semantics in AM-RADIO-Style Multi-Teacher Distillation with PHI-S and Masked Image Modelling — arXiv:2610.02903

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.02903
License (as archived): CC0-1.0
Archived: 2026-10-06T09:07:24.233Z
SHA-256 of the archived text: 4767d6b91d7d861896de1bfa1ac598e06b3280391278c3886d264cd7fedb7c49
Stable URL of this page: https://metraipolis.com/documento/dc_f04a67f25d43

How to cite

ViTok: Improving Dense Semantics in AM-RADIO-Style Multi-Teacher Distillation with PHI-S and Masked Image Modelling — arXiv:2610.02903 — arXiv (en), https://arxiv.org/abs/2610.02903 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f04a67f25d43 (Metraipolis archive dc_f04a67f25d43, sha256 4767d6b91d7d861896de1bfa1ac598e06b3280391278c3886d264cd7fedb7c49)

BibTeX

@misc{dcf04a67f25d43,
  title = {ViTok: Improving Dense Semantics in AM-RADIO-Style Multi-Teacher Distillation with PHI-S and Masked Image Modelling — arXiv:2610.02903},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.02903},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f04a67f25d43; sha256 4767d6b91d7d861896de1bfa1ac598e06b3280391278c3886d264cd7fedb7c49},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.02903. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

We study how to consolidate the current VITOK progress into a single multi-teacher distillation recipe that jointly preserves global recognition and dense semantics. Our starting point is an AM-RADIO-style student distilled from SigLIP2 and DINOv3-L, where SigLIP2 supplies strong global semantics and DINOv3-L supplies stronger dense features. The central empirical issue is that the same recipe does not optimize all objectives equally well: changes that improve ImageNet-1K kNN accuracy can still degrade ADE20K segmentation. We summarize a progression of modifications that make this trade-off more explicit and more manageable: split adaptor heads for CLS and patch tokens, asymmetric cosine/MSE losses, initialization from a DINOv3-L checkpoint, teacher reweighting, masked image modeling (MIM), and PHI-S feature balancing. The resulting model reaches 83.2 patch kNN and 85.2 CLS kNN, slightly surpassing the DINOv3-L teacher on ImageNet-1K kNN classification, while PHI-S restores ADE20K performance from 46.5/58.1 to 48.5/61.0 mIoU/mAcc, matching the teacher on this dense benchmark. We also summarize negative results: scaling distillation from ImageNet-1K to ImageNet22K does not consistently help, and naively adding extra teachers such as SAM3 or HOG features introduces interference. Rather than claiming a final recipe, this paper distills the current project state into a compact empirical story and a concrete set of lessons for future iterations.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-10-05 in cs.CV).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.02903 — 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 2026-10-06T08:21:44.182Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-10-05?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).