Metraipolis › archive › dc_3bfc016ea54e

CAST: Causal Advantage-Structured Training with Spatially Grounded Compositional Rewards for Diffusion Models — arXiv:2609.39441

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.39441
License (as archived): CC0-1.0
Archived: 2026-10-07T12:45:23.499Z
SHA-256 of the archived text: becaf329c98634af60bb1b25bdfcce0f74128b3769513c230b356a93f843c691
Stable URL of this page: https://metraipolis.com/documento/dc_3bfc016ea54e

How to cite

CAST: Causal Advantage-Structured Training with Spatially Grounded Compositional Rewards for Diffusion Models — arXiv:2609.39441 — arXiv (en), https://arxiv.org/abs/2609.39441 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3bfc016ea54e (Metraipolis archive dc_3bfc016ea54e, sha256 becaf329c98634af60bb1b25bdfcce0f74128b3769513c230b356a93f843c691)

BibTeX

@misc{dc3bfc016ea54e,
  title = {CAST: Causal Advantage-Structured Training with Spatially Grounded Compositional Rewards for Diffusion Models — arXiv:2609.39441},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.39441},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3bfc016ea54e; sha256 becaf329c98634af60bb1b25bdfcce0f74128b3769513c230b356a93f843c691},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39441. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Online reinforcement learning has been extended to flow matching for diffusion model (DM) image generation. However, this paradigm faces three limitations: (1) Window selection. Existing methods manually set the stochastic differential equation (SDE) sampling window, i.e., the denoising steps where exploration noise is injected. We instead determine it from each model's denoising trajectory. (2) Reward saturation. Current methods rely on scoring models trained on human annotations; we find that such scores are extremely high and nearly indistinguishable on the latest SOTA open-source DMs, making advantage estimation largely ineffective. (3) Sample inefficiency. A single scalar reward collapses different failure modes into almost identical scores, leaving minimal gradient guidance for targeted improvement. To address these issues, we propose CAST (Causal Advantage-Structured Training), an RL fine-tuning method for pretrained DMs, which (1) identifies the denoising step at which each model fixes the objects and their spatial arrangement in the image and uses that timing to set the SDE window, (2) decomposes each prompt via Causal Scene Graphs (CSG) into verifiable-atoms, i.e., minimal semantic units such as an object, count, attribute, or spatial relation that can each be checked independently, and rewards each atom separately, and (3) projects the signed atom-level advantages into pixel space through teacher-forced attention and uses them to spatially weight the SDE policy objective. We fine-tune two of the strongest open-source DMs, FLUX.2-dev and Qwen-Image-2512, with CAST, and evaluate them on GenEval 2, a compositional benchmark, and on Qwen-Image-Bench for overall quality. Within almost the same training budget, CAST's improvement over the base model on the most challenging GenEval 2 prompts is up to 3.07x that of Flow-GRPO, while overall generation quality also improves.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.39441 — 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 2026-10-07T12:14:14.083Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).