Metraipolis › archive › dc_96c99e4cffee
Diffusable Latents from Structure-Agnostic Distillation — arXiv:2609.39657
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.39657
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T14:46:20.041Z
- SHA-256 of the archived text
f68f35b4aa5ad595868e16d52c1ba745da04e1bcba0497b831acf9e95300e898- Stable URL of this page
https://metraipolis.com/documento/dc_96c99e4cffee
How to cite
Diffusable Latents from Structure-Agnostic Distillation — arXiv:2609.39657 — arXiv (en), https://arxiv.org/abs/2609.39657 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_96c99e4cffee (Metraipolis archive dc_96c99e4cffee, sha256 f68f35b4aa5ad595868e16d52c1ba745da04e1bcba0497b831acf9e95300e898)
BibTeX
@misc{dc96c99e4cffee,
title = {Diffusable Latents from Structure-Agnostic Distillation — arXiv:2609.39657},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.39657},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_96c99e4cffee; sha256 f68f35b4aa5ad595868e16d52c1ba745da04e1bcba0497b831acf9e95300e898},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39657. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Distilling pretrained foundation models into an autoencoder bottleneck improves latent diffusability, enabling diffusion models to converge faster and reach higher sample quality. Standard distillation aligns the latent at each position to a co-located teacher feature, tying the latent layout to the teacher's. We show this constraint is unnecessary: aligning a single pooled image-level descriptor to the teacher's performs as well as or slightly better than dense position-wise distillation. We compare first-order and relational pooled objectives across latent shapes and teacher modalities. First-order matching extends naturally to 1D token-sequence latents and across modalities, where distilling a text encoder into an image autoencoder still improves diffusability; a relational objective based only on each image's nearest neighbours improves it as well. Code and blog post are available at this https URL and this https URL . Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-10-01 in cs.CV). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV). Pagina del lavoro: https://arxiv.org/abs/2609.39657 — 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 2026-10-07T14:19:00.787Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).