Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via In-the-wild Cascading Flow Optimization
Paper
•
2502.02096
•
Published
We provide LoRA weights trained for multi-target tasks using Res152 and Incv3 as surrogate models, as well as weights further fine-tuned for 8 individual classes.
| Source | Multi-Targeted | 62 | 150 | 426 | 507 | 590 | 715 | 843 | 952 |
|---|---|---|---|---|---|---|---|---|---|
| Res152 | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA |
| Incv3 | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA |
| Source | Multi-Targeted | 62 | 150 | 426 | 507 | 590 | 715 | 843 | 952 |
|---|---|---|---|---|---|---|---|---|---|
| Res152 | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA |
| Incv3 | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA | LoRA |
If you find Dual-Flow is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.
@article{chen2025dual,
title={Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via In-the-wild Cascading Flow Optimization},
author={Chen, Yixiao and Sun, Shikun and Li, Jianshu and Li, Ruoyu and Li, Zhe and Xing, Junliang},
journal={arXiv preprint arXiv:2502.02096},
year={2025}
}
Base model
stable-diffusion-v1-5/stable-diffusion-v1-5