Scout-Refine-Correct (SRC) is a training recipe that equips flow-matching image generation models with an inference-time refinement capability. During sampling, the model performs a scout rollout to predict a full sample, uses this prediction as a self-conditioning signal, and then takes the actual step; this process can be repeated for multiple refinement rounds. On ImageNet-256, SRC improves LightningDiT-B’s FID from ~14.5 to ~7.26 with 550 normalised flow evaluations, matching a best-of‑N baseline that relies on an oracle verifier, and further reduces FID to 3.81 when combined with best-of‑N.
By Sergey Karpukhin, Aibek Alanov
arXiv:2512. 19311v2 Announce Type: replace-cross Abstract: This paper studies the training-testing discrepancy (a.
By Hui Li, Fu-Yun Wang, Haoyuan Xia, Jiayue Lyu, Kaihui Cheng, Siyu Zhu, Jingdong Wang
Preconditioned Flow Matching introduces a precondition-then-match framework that transforms target distributions into more isotropic representations before training flow matching models. By reshaping intermediate distributions, the method improves conditioning, leading to better low-eigenvalue recovery and enhanced performance metrics such as FID, MMD, precision, and recall across Gaussian, Gaussian-mixture, and high-resolution image datasets. The approach is theoretically justified and empirically validated, with ablations showing gains stem from improved geometry rather than extra parameters.
By Shadab Ahamed, Eshed Gal, Md Shahriar Rahim Siddiqui, Simon Ghyselincks, Moshe Eliasof, Eldad Haber
The study investigates whether the usefulness of training samples depends on the data itself or on the learner. By fixing selected subsets and varying the learner, the authors show that the crossover point—where easy samples give way to coverage samples—shifts with changes in model width, input resolution, and architecture. This indicates that the boundary between easy and coverage samples is learner‑dependent rather than data‑dependent.
By Yangze Liu, Xiao-Long Yin, Zhongyi Han
The paper introduces pass@k, a metric that evaluates how well generative diffusion models cover their output distribution by measuring the probability that at least one of k independent samples meets a quality criterion. Using this metric, the authors show that while classifier‑free guidance improves single‑draw quality, its advantage diminishes or reverses as k increases, revealing a trade‑off between quality and distribution coverage. They further demonstrate that the choice of training objective in diffusion distillation—whether distribution‑matching or consistency/trajectory‑based—determines whether a few‑step model retains its teacher’s coverage or sacrifices it for higher single‑draw performance, a phenomenon that also appears in few‑step causal video generation.
By Yifei Wang, Xiaoyu Wu, Tsu-Jui Fu, Chen Chen, Liang-Chieh Chen, Zhe Gan, Chen Wei
arXiv:2608. 19540v1 Announce Type: new Abstract: Training fast generators on new domains with limited data remains challenging for two reasons.
By Yara Bahram, Zahra Dehghani, M\'elodie Desbos, Eric Granger, Pablo Piantanida, Mohammadhadi Shateri