analiza aiqbench
This research introduces continuous adversarial flow models, which employ a learned discriminator to refine the training objectives of existing flow-matching architectures. By shifting from fixed mean-squared-error criteria to an adversarial approach, the method significantly enhances sample alignment and generation quality on benchmarks like ImageNet and text-to-image tasks. The work represents a notable advancement in generative modeling techniques, specifically targeting improvements in latent-space and pixel-space diffusion processes.
Continuous Adversarial Flow Models
arXiv:2604.11521v2 Announce Type: replace Abstract: We propose continuous adversarial flow models, a type of continuous-time flow model trained with an adversarial objective. Unlike flow matching, which uses a fixed mean-squared-error criterion, our approach introduces a learned discriminator to guide training. This change in objective induces a different generalized distribution, which empirically produces samples that are better aligned with the target data distribution. Our method is primarily proposed for post-training existing flow-matching models, although it can also train models from scratch. On the ImageNet 256px generation task, our post-training substantially improves the guidance-free FID of latent-space SiT from 8.26 to 3.63 and of pixel-space JiT from 7.17 to 3.57. It also improves guided generation, reducing FID from 2.06 to 1.53 for SiT and from 1.86 to 1.80 for JiT. We further evaluate our approach on text-to-image generation, where it achieves improved results on both the GenEval and DPG benchmarks.
Abstract:We propose continuous adversarial flow models, a type of continuous-time flow model trained with an adversarial objective. Unlike flow matching, which uses a fixed mean-squared-error criterion, our approach introduces a learned discriminator to guide training. This change in objective induces a different generalized distribution, which empirically produces samples that are better aligned with the target data distribution. Our method is primarily proposed for post-training existing flow-matching models, although it can also train models from scratch. On the ImageNet 256px generation task, our post-training substantially improves the guidance-free FID of latent-space SiT from 8.26 to 3.63 and of pixel-space JiT from 7.17 to 3.57. It also improves guided generation, reducing FID from 2.06 to 1.53 for SiT and from 1.86 to 1.80 for JiT. We further evaluate our approach on text-to-image generation, where it achieves improved results on both the GenEval and DPG benchmarks.
Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2604.11521 [cs.LG]
(or arXiv:2604.11521v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2604.11521 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 241 cuvinte
Subiecte conexe
Continuous Adversarial MeanFlow Transfer
arXiv · 21 aug. 2026
Netskope (NTSK) Q2 2027 Earnings Call Transcript
fool.com · 10 sept. 2026
Rethinking Robot Safety in the Age of AI
IEEE Spectrum · 16 sept. 2026
Against Export Controls (and China Threat Models)
LessWrong · 24 sept. 2026