Commit
β’
4460b58
1
Parent(s):
917bc12
readme init
Browse files
README.md
CHANGED
@@ -1,3 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
-
---
|
|
|
1 |
+
|
2 |
+
|
3 |
+
# Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
|
4 |
+
### [Project Website]() | [Arxiv Preprint](https://arxiv.org/abs/2405.15234) | [Fine-tuned Weights](https://drive.google.com/drive/folders/1Nf-EJ2W3CsZwpc5blZFi7tm7o1wEiTg4?usp=sharing) | [Demo]() <br>
|
5 |
+
Our proposed robust unlearning framework, AdvUnlearn, enhances diffusion models' safety by robustly erasing unwanted concepts through adversarial training, achieving an optimal balance between concept erasure and image generation quality.
|
6 |
+
|
7 |
+
|
8 |
+
|
9 |
+
|
10 |
+
|
11 |
+
## Baselines
|
12 |
+
|
13 |
+
| DM Unlearning Methods | Nudity | Van Gogh | Objects |
|
14 |
+
|:-------|:----:|:-------:| :-------:|
|
15 |
+
| [ESD](https://github.com/rohitgandikota/erasing) (Erased Stable Diffusion) | β
| β
| β
|
16 |
+
| [FMN](https://github.com/SHI-Labs/Forget-Me-Not) (Forget-Me-Not) | β
| β
| β
|
17 |
+
| [AC](https://github.com/nupurkmr9/concept-ablation) (Ablating Concepts) | β | β
| β
|
18 |
+
| [UCE](https://github.com/rohitgandikota/unified-concept-editing) (Unified Concept Editing) | β
| β
| β
|
19 |
+
| [SalUn](https://github.com/OPTML-Group/Unlearn-Saliency) (Saliency Unlearning) | β
| β | β
|
20 |
+
| [SH](https://github.com/JingWu321/Scissorhands_ex) (ScissorHands) | β
| β | β
|
21 |
+
| [ED](https://github.com/JingWu321/EraseDiff) (EraseDiff) | β
| β | β
|
22 |
+
| [SPM](https://github.com/Con6924/SPM) (concept-SemiPermeable Membrane) | β
| β
| β
|
23 |
+
| **AdvUnlearn (Ours)** | β
| β
| β
|
24 |
+
|
25 |
+
|
26 |
+
<br>
|
27 |
+
|
28 |
+
|
29 |
+
## Cite Our Work
|
30 |
+
The preprint can be cited as follows:
|
31 |
+
```
|
32 |
+
@misc{zhang2024defensive,
|
33 |
+
title={Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models},
|
34 |
+
author={Yimeng Zhang and Xin Chen and Jinghan Jia and Yihua Zhang and Chongyu Fan and Jiancheng Liu and Mingyi Hong and Ke Ding and Sijia Liu},
|
35 |
+
year={2024},
|
36 |
+
eprint={2405.15234},
|
37 |
+
archivePrefix={arXiv},
|
38 |
+
primaryClass={cs.CV}
|
39 |
+
}
|
40 |
+
```
|
41 |
+
|
42 |
---
|
43 |
license: cc-by-4.0
|
44 |
+
---
|