-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
342 lines (315 loc) · 16.5 KB
/
index.html
File metadata and controls
342 lines (315 loc) · 16.5 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description"
content="RAR unifies image restoration and quality assessment into a single latent iterative framework.
Instead of treating restoration and evaluation as separate stages, RAR aligns a free-form vision-language quality model with a flow-matching generative prior in latent space.
Through an end-to-end Restore–Assess–Repeat loop, the model progressively refines degraded inputs under unknown and composite distortions, achieving both perceptual improvement and adaptive stopping without external supervision.">
<meta property="og:title" content="Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration" />
<meta property="og:description" content="RAR is an iterative image restoration framework that unifies assessment and restoration in a shared latent space." />
<meta property="og:url" content="URL OF THE WEBSITE" />
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content="static/image/your_banner_image.png" />
<meta property="og:image:width" content="1200" />
<meta property="og:image:height" content="630" />
<meta name="twitter:title" content="Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration">
<meta name="twitter:description" content="RAR is an iterative image restoration framework that unifies assessment and restoration in a shared latent space.">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image" content="static/images/your_twitter_banner_image.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="Image Restoration, Image Quality Assessment, Diffusion Models, Flow Matching">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration
</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="image_src" href="./rar_files/logo.png">
<link rel="icon"
type="image/x-icon"
href="./rar_files/logo.png"/>
<link rel="stylesheet" href="static/css/bulma.min.css">
<link rel="stylesheet" href="static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="static/css/bulma-slider.min.css">
<link rel="stylesheet" href="static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static/js/fontawesome.all.min.js"></script>
<script src="static/js/bulma-carousel.min.js"></script>
<script src="static/js/bulma-slider.min.js"></script>
<script src="static/js/index.js"></script>
</head>
<body>
<!-- Authors -->
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-1 publication-title">
Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration
</h2>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<div class="is-size-5 publication-authors">
<div class="author-block"><a href="https://sites.google.com/view/cihsiang/home">I-Hsiang Chen</a> <sup> 1,2 </sup></div>
<div class="author-block"><a href="">Isma Hadji</a> <sup> 1 </sup></div>
<div class="author-block"><a href="">Enrique Sanchez</a> <sup> 1 </sup></div>
<div class="author-block"><a href="">Adrian Bulat</a> <sup> 1,3 </sup></div>
<div class="author-block"><a href="https://homepage.ntu.edu.tw/~sykuo">Sy-Yen Kuo</a> <sup> 2,4 </sup></div>
<div class="author-block"><a href="">Radu Timofte</a> <sup> 5 </sup></div>
<div class="author-block"><a href="">Georgios Tzimiropoulos</a> <sup> 1,6 </sup></div>
<div class="author-block"><a href="">Brais Martinez</a> <sup> 1 </sup>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Samsung AI Center Cambridge</span>
<span class="author-block"><sup>2</sup>National Taiwan University</span>
<span class="author-block"><sup>3</sup>Technical University of Iasi</span>
<span class="author-block"><sup>4</sup>Chang Gung University</span>
<span class="author-block"><sup>5</sup>University of Wurzburg</span>
<span class="author-block"><sup>6</sup>Queen Mary University of London</spa>
</div>
<!-- <span class="eql-cntrb"><sup>*</sup>Equal Contribution <sup>✝</sup>Corresponding
Author</span> -->
<div class="is-size-5 publication-authors">
<span class="author-block"> </sup>(CVPR 2026)</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2603.26385"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<span class="link-block">
<a href="https://arxiv.org/abs/2603.26385"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<span class="link-block">
<a href="https://github.com/saic-fi/RAR"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- Dataset Link. -->
<span class="link-block">
<a href="https://github.com/saic-fi/RAR"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="far fa-images"></i>
</span>
<span>Data</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Video -->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<video autoplay muted loop playsinline style="width: 100%; height: auto; border-radius: 12px;">
<source src="./rar_files/Teaser.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
<h2 class="subtitle has-text-centered">
<span class="methodname">RAR</span> is an iterative image restoration framework that unifies assessment and restoration in a shared latent space.
</h2>
</div>
</div>
</section>
<!-- Abstract -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h4 class="title is-3">Abstract</h4>
<div class="content has-text-justified">
<p>
Image restoration aims to recover high quality images from inputs degraded by various factors, such as adverse weather, blur, or low light. While recent studies have shown remarkable progress across individual or unified restoration tasks, they still suffer from limited generalization and inefficiency when handling unknown or composite degradations. To address these limitations, we propose RAR, a Restore, Assess and Repeat process, that integrates Image Quality Assessment (IQA) and Image Restoration (IR) into a unified framework to iteratively and efficiently achieve high quality image restoration. Specifically, we introduce a restoration process that operates entirely in the latent domain to jointly perform degradation identification, image restoration, and quality verification. The resulting model is fully trainable end to end and allows for an all-in-one assess and restore approach that dynamically adapts the restoration process. Also, the tight integration of IQA and IR into a unified model minimizes the latency and information loss that typically arises from keeping the two modules disjoint, (e.g. during image and/or text decoding). Extensive experiments show that our approach consistent improvements under single, unknown and composite degradations, thereby establishing a new state-of-the-art.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- Method -->
<section class="section">
<div class="container is-max-desktop">
<div class="content has-text-justified">
<h4 class="title is-3 has-text-centered">Architecture of RAR</h4>
<div class="has-text-centered">
<div class="content has-text-justified">
<p>
RAR consists of two tightly integrated modules:
<span class="tag is-light" style="background-color: #e7f6e7; color: #1e5e1e;"><b>Latent Quality Assessment (LQA)</b></span> and
<span class="tag is-light" style="background-color: #f8e6da; color: #a25803;"><b>Unified Image Restoration (UIR)</b></span>.
By operating entirely in the shared latent space, RAR enables iterative degradation identification, restoration, and verification in a unified end-to-end framework.
</p>
</div>
<img style="width: 100%;" src="./rar_files/model.png"
alt="Overview of RAR."/>
</div>
<div class="columns is-multiline is-variable is-6">
<!-- LQA -->
<div class="column is-half">
<p class="content has-text-justified">
<span class="tag is-light" style="background-color: #e7f6e7; color: #1e5e1e;"><b>Latent Quality Assessment (LQA)</b></span>
projects degraded image latents into the IQA space via an image adapter, and aligns IQA outputs directly to restoration conditioning embeddings through a text adapter.
This removes lossy text decoding and enables end-to-end differentiable feedback between assessment and restoration.
</p>
</div>
<!-- UIR -->
<div class="column is-half">
<p class="content has-text-justified">
<span class="tag is-light" style="background-color: #f8e6da; color: #a25803;"><b>Unified Image Restoration (UIR)</b></span>
is a flow-matching–based generative restoration backbone that directly maps degraded latents to high-quality latents.
Its noise-free formulation allows intermediate latent reassessment, enabling stable multi-round restoration.
</p>
</div>
</div>
<p>
<em>
Together, LQA and UIR form a closed-loop latent feedback system that enables adaptive Restore–Assess–Repeat iterations with a quality-aware stopping criterion.
</em>
</p>
</div>
</section>
<!-- Effectiveness of Proposed Modules -->
<section class="hero is-light">
<div class="hero-body">
<div class="container is-max-desktop">
<h4 class="title is-3 has-text-centered">Effectiveness of Proposed Modules</h4>
<div id="results-carousel" class="carousel results-carousel">
<!-- Sample 1 -->
<div class="item">
<div>
<img src="rar_files/ablation/noise-free.png"/>
</div>
<h5 class="subtitle is-6 has-text-centered result-caption">
Noise-free flow matching enables stable iterative conditioning, preventing latent corruption and improving perceptual quality across degradations.
</h5>
</div>
<!-- Sample 2 -->
<div class="item">
<div>
<img src="rar_files/ablation/embedding.png"/>
</div>
<h5 class="subtitle is-6 has-text-centered result-caption">
Replacing text decoding with direct embedding alignment preserves richer degradation cues and strengthens restoration conditioning.
</h5>
</div>
<!-- Sample 3 -->
<div class="item">
<div>
<img src="rar_files/ablation/latent.png"/>
</div>
<h5 class="subtitle is-6 has-text-centered result-caption">
Aligning IQA and restoration in a shared latent space eliminates encoding/decoding information loss and enables tightly coupled assess–restore feedback.
</h5>
</div>
<!-- Sample 4 -->
<div class="item">
<div>
<img src="rar_files/ablation/ablation.png"/>
</div>
<h5 class="subtitle is-6 has-text-centered result-caption">
Our unified design delivers robust gains across architectures and datasets, establishing strong generalization under unknown and composite degradations.
</h5>
</div>
</div>
</div>
</div>
</section>
<!-- Qualitative Results -->
<section class="section">
<div class="hero-body">
<div class="container is-max-desktop">
<h4 class="title is-3 has-text-centered">Qualitative Results</h4>
<div class="content has-text-justified">
<p>
RAR performs restoration through iterative assess–restore steps.
For each example, we visualize the progressive improvement across multiple rounds, demonstrating how degradations are identified and removed sequentially.
</p>
</div>
<div id="results-carousel" class="carousel results-carousel">
<div class="item">
<img src="rar_files/results/Sample5.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample6.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample7.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample8.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample1.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample2.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample3.png"/>
</div>
<div class="item">
<img src="rar_files/results/Sample4.png"/>
</div>
</div>
</div>
</div>
</section>
<hr/>
<!-- <section class="section" id="BibTeX">
<div class="container content is-max-desktop">
<h2 class="title">BibTeX</h2>
<pre><code>@misc{chen2026robustvisragcausalityawarevisionbasedretrievalaugmented,
title={RobustVisRAG: Causality-Aware Vision-Based Retrieval-Augmented Generation under Visual Degradations},
author={I-Hsiang Chen and Yu-Wei Liu and Tse-Yu Wu and Yu-Chien Chiang and Jen-Chien Yang and Wei-Ting Chen},
year={2026},
eprint={2602.22013},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2602.22013},
}
</code></pre>
</div>
</section> -->
<!-- Footer -->
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link" href="">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="" class="external-link" disabled>
<i class="fab fa-github"></i>
</a>
</div>
</div>
</footer>
</body>
</html>