Metraipolis › archive › dc_cd2c4ce5ee7e

Scout-Refine-Correct: Teaching Flow Models to Refine Their Own Samples — arXiv:2610.41598

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.41598
License (as archived): CC0-1.0
Archived: 2026-11-02T08:03:23.102Z
SHA-256 of the archived text: 9c3fae6ee51ec84b3b54a5e8b7c505fde778df72ef3a5759df1d26104f00e30e
Stable URL of this page: https://metraipolis.com/documento/dc_cd2c4ce5ee7e

How to cite

Scout-Refine-Correct: Teaching Flow Models to Refine Their Own Samples — arXiv:2610.41598 — arXiv (en), https://arxiv.org/abs/2610.41598 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cd2c4ce5ee7e (Metraipolis archive dc_cd2c4ce5ee7e, sha256 9c3fae6ee51ec84b3b54a5e8b7c505fde778df72ef3a5759df1d26104f00e30e)

BibTeX

@misc{dccd2c4ce5ee7e,
  title = {Scout-Refine-Correct: Teaching Flow Models to Refine Their Own Samples — arXiv:2610.41598},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.41598},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cd2c4ce5ee7e; sha256 9c3fae6ee51ec84b3b54a5e8b7c505fde778df72ef3a5759df1d26104f00e30e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.41598. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

The quality of image generation models trained with diffusion or flow matching is largely fixed by the training budget: at inference, the only available lever is the number of sampling steps, and additional steps improve quality only up to a point. We present a training recipe, Scout-Refine-Correct (SRC), that instead gives a flow-matching generator a genuine inference-time refinement capability. At each sampling step the model takes a scout rollout -- $K$ steps ahead along its own trajectory -- extrapolates a full-sample estimate, and feeds that estimate back to itself as a self-conditioning signal before taking the real step. This can be repeated for $R$ refinement rounds. Two correction losses, applied during training at states the model's own rollouts actually reach rather than only on the ground-truth path, teach the model to use this signal to reach new samples instead of collapsing to a suboptimal one. On ImageNet-256, SRC improves LightningDiT-B from FID~14.5 to FID~7.26 at NFE~550, matching a privileged best-of-$N$ baseline that uses an oracle FID verifier to choose among samples. SRC uses no external verifier, no reward signal, and, after ablation, almost no extra parameters over the base architecture. Combining SRC with best-of-$N$ improves the result further, to FID~3.81 at NFE~4400, showing that the two axes are complementary. We ablate the main design choices -- the architectural pathway for the scout signal, the scout horizon, each correction loss, and the training schedule to demonstrate key working components of training.

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/2610.41598 — 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 2026-11-02T07:44:00.758Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).