Metraipolis › archive › dc_40a46abc8db2

Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts — arXiv:2609.35641

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.35641
License (as archived)
CC0-1.0
Archived
2026-10-07T20:59:00.204Z
SHA-256 of the archived text
b1590d06bbf09f6d1560ea49ee3826bfc05f8910189e46f2f4266054493c7d2f
Stable URL of this page
https://metraipolis.com/documento/dc_40a46abc8db2

How to cite

Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts — arXiv:2609.35641 — arXiv (en), https://arxiv.org/abs/2609.35641 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_40a46abc8db2 (Metraipolis archive dc_40a46abc8db2, sha256 b1590d06bbf09f6d1560ea49ee3826bfc05f8910189e46f2f4266054493c7d2f)
BibTeX
@misc{dc40a46abc8db2,
  title = {Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts — arXiv:2609.35641},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.35641},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_40a46abc8db2; sha256 b1590d06bbf09f6d1560ea49ee3826bfc05f8910189e46f2f4266054493c7d2f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35641. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Precise instruction following in image generation, such as satisfying object counts and spatial relations, remains an open challenge at least in part because it is learned using unreliable reward models such as object detectors and vision-language models. We introduce Verifiable Visual Rewards (VVR), the first framework for programmatically verifiable image rewards, and show that training on it generalizes to natural prompts. Each VVR task is a scene of geometric objects and relations among them, from which we derive both the prompt and a deterministic verifier, so tasks can be generated in any number and at any chosen complexity. We release VVRBench, with 10,000 tasks over 32 constraint types, and VVRBench-Challenge, with 720 more complex tasks; the strongest model we evaluate---GPT-Image-2.5---solves 21.4% of VVRBench-Challenge. Using VVR scores as rewards for reinforcement learning (RLVVR) raises the accuracy of Stable Diffusion 3.5 Medium on VVRBench from 2.8% to 28.3% and demonstrates consistent easy-to-hard generalization. These gains extend to out-of-domain benchmarks, and mixing VVR into existing objectives further improves overall performance and human preference, motivating the adoption of VVR into standard image generation post-training recipes.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2609.35641 — 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 2026-10-07T20:29:01.918Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).