Skip to content

Commit cb2673b

Browse files
committed
DiffVax ICLR 2026 update
1 parent 96d368a commit cb2673b

9 files changed

Lines changed: 56 additions & 25 deletions

File tree

.DS_Store

6 KB
Binary file not shown.

index.html

Lines changed: 56 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@
55
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
66
<!-- Replace the content tag with appropriate information -->
77
<meta name="description" content="DiffVax is a scalable, lightweight, and fast framework for image and video immunization, preventing diffusion-based editing with imperceptible noise and achieving unprecedented speed and robustness.">
8-
<meta property="og:title" content="Optimization-Free Image Immunization Against Diffusion-Based Editing"/>
8+
<meta property="og:title" content="DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing"/>
99
<meta property="og:description" content="DiffVax is a scalable, lightweight, and fast framework for image and video immunization, preventing diffusion-based editing with imperceptible noise and achieving unprecedented speed and robustness."/>
1010
<meta property="og:url" content="https://diffvax.github.io/"/>
1111
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
@@ -14,7 +14,7 @@
1414
<meta property="og:image:height" content="630"/>
1515

1616

17-
<meta name="twitter:title" content="Optimization-Free Image Immunization Against Diffusion-Based Editing">
17+
<meta name="twitter:title" content="DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing">
1818
<meta name="twitter:description" content="DiffVax is a scalable, lightweight, and fast framework for image and video immunization, preventing diffusion-based editing with imperceptible noise and achieving unprecedented speed and robustness.">
1919
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
2020
<meta name="twitter:image" content="static/images/your_twitter_banner_image.png">
@@ -24,7 +24,7 @@
2424
<meta name="viewport" content="width=device-width, initial-scale=1">
2525

2626

27-
<title>Optimization-Free Image Immunization Against Diffusion-Based Editing</title>
27+
<title>DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing</title>
2828
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
2929
rel="stylesheet">
3030

@@ -64,15 +64,16 @@
6464
<div class="container is-max-desktop">
6565
<div class="columns is-centered">
6666
<div class="column has-text-centered">
67-
<h1 class="xtitle is-1 publication-title" style=" font-size: 200%; font-weight: 800;">Optimization-Free Image Immunization Against Diffusion-Based Editing</h1>
67+
<h1 class="xtitle is-1 publication-title" style=" font-size: 200%; font-weight: 800;">DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing</h1>
68+
<p class="is-size-4 publication-title" style="margin-top: 0.5rem; margin-bottom: 1rem;">ICLR 2026</p>
6869
<div class="is-size-5 publication-authors">
6970
<!-- Paper authors -->
70-
<span class="author-block"><a href="https://www.linkedin.com/in/tarikcanozden/" target="_blank">Tarik Can Ozden</a><sup>1,2,3*†</sup></span>
71+
<span class="author-block"><a href="https://www.linkedin.com/in/tarikcanozden/" target="_blank">Tarik Can Ozden</a><sup>1*</sup></span>
7172
<span class="author-block"><a href="https://karaozgur.com/" target="_blank">Ozgur Kara</a><sup>1*</sup></span>
7273
<span class="author-block"><a href="https://oguzhanakcin.github.io/" target="_blank">Oguzhan Akcin</a><sup>2</sup></span>
73-
<span class="author-block"><a href="https://keremzaman.com/" target="_blank">Kerem Zaman</a><sup>4</sup></span>
74+
<span class="author-block"><a href="https://keremzaman.com/" target="_blank">Kerem Zaman</a><sup>3</sup></span>
7475
<br>
75-
<span class="author-block"><a href="https://www.ssriva.com/" target="_blank">Shashank Srivastava</a><sup>4</sup></span>
76+
<span class="author-block"><a href="https://www.ssriva.com/" target="_blank">Shashank Srivastava</a><sup>3</sup></span>
7677
<span class="author-block"><a href="https://scholar.google.com/citations?user=262ASa4AAAAJ&hl=en" target="_blank">Sandeep P. Chinchali</a><sup>2</sup></span>
7778
<span class="author-block"><a href="https://scholar.google.com/citations?hl=en&user=8kA3eDwAAAAJ" target="_blank">James M. Rehg</a><sup>1</sup></span>
7879
</div>
@@ -81,22 +82,20 @@ <h1 class="xtitle is-1 publication-title" style=" font-size: 200%; font-weigh
8182
<span class="author-block"><sup>1</sup>University of Illinois Urbana-Champaign</span>
8283
<span class="author-block"><sup>2</sup>The University of Texas at Austin</span>
8384
<br>
84-
<span class="author-block"><sup>3</sup>Bogazici University</span>
85-
<span class="author-block"><sup>4</sup>University of North Carolina at Chapel Hill</span>
85+
<span class="author-block"><sup>3</sup>University of North Carolina at Chapel Hill</span>
8686
<span class="eql-cntrb"><small><br> <sup>*</sup>Equal contribution</small></span>
87-
<span class="eql-cntrb"><small><br> <sup></sup>Work done during an internship at UT Austin and UIUC</small></span>
8887
</div>
8988

9089
<div class="column has-text-centered">
9190
<div class="publication-links">
9291
<!-- Arxiv PDF link -->
9392
<span class="link-block">
94-
<a href="https://arxiv.org/abs/2411.17957" target="_blank"
93+
<a href="static/pdfs/DiffVax_ICLR2026.pdf" target="_blank"
9594
class="external-link button is-normal is-rounded is-dark">
9695
<span class="icon">
97-
<i class="ai ai-arxiv"></i>
96+
<i class="fas fa-file-pdf"></i>
9897
</span>
99-
<span>arXiv</span>
98+
<span>PDF</span>
10099
</a>
101100
</span>
102101

@@ -150,7 +149,7 @@ <h2 class="subtitle has-text-centered">
150149
<h2 class="title is-3 hr-lines">Abstract</h2>
151150
<div class="content has-text-justified">
152151
<p>
153-
Current image immunization defense techniques against diffusion-based editing embed imperceptible noise in target images to disrupt editing models. However, these methods face scalability challenges, as they require time-consuming re-optimization for each image—taking hours for small batches. To address these challenges, we introduce DiffVax, a scalable, lightweight, and optimization-free framework for image immunization, specifically designed to prevent diffusion-based editing. Our approach enables effective generalization to unseen content, reducing computational costs and cutting immunization time from days to milliseconds—achieving a 250,000$\times$ speedup. This is achieved through a loss term that ensures the failure of editing attempts and the imperceptibility of the perturbations. Extensive qualitative and quantitative results demonstrate that our model is scalable, optimization-free, adaptable to various diffusion-based editing tools, robust against counter-attacks, and, for the first time, effectively protects video content from editing. Our code and qualitative results are provided in the supplementary.
152+
Current image immunization defense techniques against diffusion-based editing embed imperceptible noise into target images to disrupt editing models. However, these methods face scalability challenges, as they require time-consuming optimization for each image separately, taking hours for small batches. To address these challenges, we introduce DiffVax, a scalable, lightweight, and optimization-free framework for image immunization, specifically designed to prevent diffusion-based editing. Our approach enables effective generalization to unseen content, reducing computational costs and cutting immunization time from days to milliseconds, achieving a speedup of 250,000×. This is achieved through a loss term that ensures the failure of editing attempts and the imperceptibility of the perturbations. Extensive qualitative and quantitative results demonstrate that our model is scalable, optimization-free, adaptable to various diffusion-based editing tools, robust against counter-attacks, and, for the first time, effectively protects video content from editing.
154153
</p>
155154
</div>
156155
</div>
@@ -160,13 +159,42 @@ <h2 class="title is-3 hr-lines">Abstract</h2>
160159
<!-- End paper abstract -->
161160

162161

163-
<!-- Paper poster -->
162+
<!-- Overview -->
164163
<section class="hero is-small">
165164
<div class="hero-body">
166165
<div class="container is-max-desktop">
167166
<div class="columns is-centered has-text-centered">
168167
<div class="column is-">
169-
<h2 class="title is-3 hr-lines">Method</h2>
168+
<h2 class="title is-3 hr-lines">Overview</h2>
169+
<div class="content has-text-justified">
170+
171+
<table width="800" border="0">
172+
<tbody>
173+
<tr>
174+
<td colspan="3">
175+
<p>
176+
<strong>(a)</strong> An attacker performs malicious editing on an original image.
177+
178+
<strong>(b)</strong> Existing defenses immunize images by solving a costly optimization problem for each image individually, taking over 10 minutes per image.
179+
<strong>(c)</strong> DiffVax enables scalable protection by first training an immunizer model (green box) on a diverse dataset. Once trained, the model can immunize unseen images with a single forward pass, producing effective perturbations in approximately 70 milliseconds per image. </p>
180+
</td>
181+
</tr>
182+
<img src="static/images/FIG_motivation.png" alt="overview"/>
183+
</tbody>
184+
</table>
185+
</div>
186+
</div>
187+
</section>
188+
<!--End Overview -->
189+
190+
191+
<!-- Paper poster -->
192+
<section class="hero is-light">
193+
<div class="hero-body">
194+
<div class="container is-max-desktop">
195+
<div class="columns is-centered has-text-centered">
196+
<div class="column is-">
197+
<h2 class="title is-3 hr-lines">Methodology</h2>
170198
<div class="content has-text-justified">
171199

172200
<table width="800" border="0">
@@ -177,7 +205,10 @@ <h2 class="title is-3 hr-lines">Method</h2>
177205
Our process begins with the immunizer model $f(\cdot;\theta)$ which generates imperceptible noise $\epsilon_{im}$ to be applied to original image $I$. This noise is applied to the masked region $M$ of the image, resulting in immunized image $I_{im}$. The immunized image is then processed by a diffusion-based editing model $SD(\cdot)$ using a text prompt $P$ and the complementary mask $\sim M$ to edit the background of the original image. The training aims to minimize two loss terms $\mathcal{L}_{noise}$ and $\mathcal{L}_{edit}$, which penalizes the applied noise magnitude, and if the edit is successful, respectively. During training, the immunizer learns to generalize across diverse images, ensuring editing attempts fail while preserving visual fidelity. This end-to-end framework enables robust, scalable immunization against diffusion-based editing for both images and videos.</p>
178206
</td>
179207
</tr>
180-
<img src="static/images/immunuzation.png" alt="model_overview"/>
208+
<img src="static/images/FIG3.png" alt="model_overview"/>
209+
210+
211+
181212
</tbody>
182213
</table>
183214
</div>
@@ -187,7 +218,7 @@ <h2 class="title is-3 hr-lines">Method</h2>
187218

188219

189220
<!-- Immunization Results-->
190-
<section class="hero is-light">
221+
<section class="hero is-small">
191222
<div class="hero-body">
192223
<div class="container is-max-desktop">
193224
<div class="columns is-centered has-text-centered">
@@ -397,7 +428,7 @@ <h2 class="title is-3 hr-lines">DiffVax Immunization Results</h2>
397428

398429

399430
<!-- Comparison -->
400-
<section class="hero is-small">
431+
<section class="hero is-light">
401432
<div class="hero-body">
402433
<div class="container is-max-desktop">
403434
<div class="columns is-centered has-text-centered">
@@ -488,7 +519,7 @@ <h2 class="title is-3 hr-lines">Comparisons</h2>
488519
<!-- End Comparison -->
489520

490521
<!-- Robust against counter attacks-->
491-
<section class="hero is-light">
522+
<section class="hero is-small">
492523
<div class="hero-body">
493524
<div class="container is-max-desktop">
494525
<div class="columns is-centered has-text-centered">
@@ -556,11 +587,11 @@ <h2 class="title is-3 hr-lines">Robustness Against Counter Attacks</h2>
556587
<section class="section" id="BibTeX">
557588
<div class="container is-max-desktop content">
558589
<h2 class="title">BibTeX</h2>
559-
<pre><code>@article{ozden2024optimization,
560-
title={Optimization-Free Image Immunization Against Diffusion-Based Editing},
561-
author={Ozden, Tarik Can and Kara, Ozgur and Akcin, Oguzhan and Zaman, Kerem and Srivastava, Shashank and Chinchali, Sandeep P and Rehg, James M},
562-
journal={arXiv preprint arXiv:2411.17957},
563-
year={2024}
590+
<pre><code>@inproceedings{ozden2026diffvax,
591+
title={DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing},
592+
author={Ozden, Tarik Can and Kara, Ozgur and Akcin, Oguzhan and Zaman, Kerem and Srivastava, Shashank and Chinchali, Sandeep P and Rehg, James M},
593+
booktitle={The Fourteenth International Conference on Learning Representations},
594+
year={2026},
564595
}</code></pre>
565596
</div>
566597
</section>

static/.DS_Store

6 KB
Binary file not shown.

static/images/FIG3.png

2.11 MB
Loading

static/images/FIG_motivation.png

1.7 MB
Loading

static/images/immunuzation.png

-1.67 MB
Binary file not shown.

static/pdfs/.DS_Store

6 KB
Binary file not shown.

static/pdfs/DiffVax.pdf

-6.73 MB
Binary file not shown.

static/pdfs/DiffVax_ICLR2026.pdf

39.6 MB
Binary file not shown.

0 commit comments

Comments
 (0)