-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
166 lines (143 loc) · 7.74 KB
/
index.html
File metadata and controls
166 lines (143 loc) · 7.74 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Unified Thinker | Compact & Vibrant</title>
<link href="https://fonts.googleapis.com/css2?family=Outfit:wght@400;700;800&family=Inter:wght@400;600&display=swap" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0/css/all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<style>
:root {
--primary-gradient: linear-gradient(90deg, #4f46e5 0%, #9333ea 100%);
--accent-soft: #f5f3ff;
--text-dark: #0f172a;
--text-muted: #475569;
}
body { font-family: 'Inter', sans-serif; color: var(--text-dark); margin: 0; line-height: 1.5; background: #fff; }
.container { max-width: 950px; margin: 0 auto; padding: 10px 20px; }
/* Hero Section */
.hero { text-align: center; padding: 40px 0 20px 0; }
.title {
font-family: 'Outfit', sans-serif; font-size: 2.4rem; font-weight: 800; margin: 0 0 10px 0;
background: var(--primary-gradient); -webkit-background-clip: text; -webkit-text-fill-color: transparent;
}
.authors { font-size: 1.05rem; margin-bottom: 5px; }
.author-block { display: inline-block; margin: 0 5px; }
.institution { font-size: 0.95rem; color: var(--text-muted); margin-bottom: 15px; }
/* Buttons */
.link-container { display: flex; justify-content: center; gap: 10px; margin-bottom: 20px; }
.button {
background: #0f172a; color: #fff; border-radius: 8px; padding: 8px 18px;
text-decoration: none; font-weight: 600; font-size: 0.9rem; transition: 0.2s;
}
.button.special { background: var(--primary-gradient); }
.button:hover { transform: translateY(-2px); opacity: 0.9; }
/* Section Control */
.section { padding: 15px 0; }
.section-title {
font-family: 'Outfit', sans-serif; font-size: 1.5rem; font-weight: 700;
margin: 0 0 15px 0; padding-left: 12px; border-left: 4px solid #6366f1;
display: flex; align-items: center; color: #1e1b4b;
}
.abstract-box {
background: var(--accent-soft); padding: 20px; border-radius: 12px;
text-align: justify; font-size: 1rem; border: 1px solid #e0e7ff;
}
/* Keywords Color */
.highlight { color: #6366f1; font-weight: 700; }
/* Figures */
.figure-box { text-align: center; margin: 10px 0; }
.figure-img { max-width: 100%; border-radius: 8px; box-shadow: 0 10px 15px -3px rgba(0,0,0,0.05); }
.caption { font-size: 0.88rem; color: var(--text-muted); margin-top: 8px; padding: 0 5%; line-height: 1.3; }
/* BibTeX */
.bibtex-box {
background: #1e293b; color: #cbd5e1; padding: 15px; border-radius: 8px;
font-family: 'monospace'; font-size: 0.8rem; overflow-x: auto;
}
footer { text-align: center; padding: 30px 0; color: #94a3b8; font-size: 0.8rem; border-top: 1px solid #f1f5f9; }
@media (max-width: 768px) { .title { font-size: 1.8rem; } }
</style>
</head>
<body>
<div class="container">
<section class="hero">
<h1 class="title">Unified Thinker</h1>
<p style="font-weight: 600; color: #6366f1; margin: -5px 0 15px 0; font-size: 1.5rem;">
A General Reasoning Modular Core for Image Generation
</p>
<div class="authors">
<span class="author-block"><strong>Sashuai Zhou</strong><sup>1,2*</sup></span>
<span class="author-block"><strong>Qiang Zhou</strong><sup>2*</sup></span>
<span class="author-block"><strong>Jijin Hu</strong><sup>2*</sup></span>
<span class="author-block"><strong>Hanqing Yang</strong><sup>2*</sup></span>
<span class="author-block"><strong>Yue Cao</strong><sup>3</sup></span>
<span class="author-block"><strong>Junpeng Ma</strong><sup>4</sup></span>
<br>
<span class="author-block"><strong>Yinchao Ma</strong><sup>2</sup></span>
<span class="author-block"><strong>Jun Song</strong><sup>2†</sup></span>
<span class="author-block"><strong>Tiezheng Ge</strong><sup>2</sup></span>
<span class="author-block"><strong>Cheng Yu</strong><sup>2</sup></span>
<span class="author-block"><strong>Bo Zheng</strong><sup>2</sup></span>
<span class="author-block"><strong>Zhou Zhao</strong><sup>1†</sup></span>
</div>
<div class="institution">
<sup>1</sup>Zhejiang University •
<sup>2</sup>Alibaba Group •
<sup>3</sup>Nanjing University •
<sup>4</sup>Fudan University
</div>
<div class="link-container">
<a href="./" class="button"><i class="fas fa-home"></i> Project</a>
<a href="https://arxiv.org/abs/2601.03127" class="button"><i class="ai ai-arxiv"></i> arXiv</a>
<a href="https://github.com/LivingFutureLab/UnifiedThinker" class="button special"><i class="fab fa-github"></i> Code</a>
<a href="#" class="button"><i class="fas fa-database"></i> Dataset</a>
</div>
</section>
<div class="figure-box" style="padding: 0 20px;">
<img src="src/case.jpg" class="figure-img" alt="Main Results">
<p class="caption"><span class="highlight">Unified Thinker</span> bridges the reasoning-execution gap with modular upgrades.</p>
</div>
<section class="section">
<h2 class="section-title">Abstract</h2>
<div class="abstract-box">
Despite impressive progress in high-fidelity image synthesis, generative models still struggle with logic-intensive instruction following, exposing a persistent <span class="highlight">reasoning-execution gap</span>.
We propose <strong>Unified Thinker</strong>, a task-agnostic reasoning architecture designed as a <strong>unified planning core</strong> that can plug into diverse generators and workflows.
Unified Thinker <strong>decouples</strong> a dedicated <span class="highlight">Thinker</span> from the image <strong>Generator</strong>, enabling <strong>modular upgrades</strong> of reasoning without retraining the entire model.
We further introduce a <strong>two-stage training paradigm</strong>: first building a structured planning interface, then applying reinforcement learning to ground its policy in <strong>pixel-level feedback</strong>, ensuring plans optimize <strong>visual correctness</strong> over textual plausibility.
</div>
</section>
<section class="section">
<h2 class="section-title">Methodology</h2>
<div class="abstract-box" style="background: transparent; border: none; padding: 0 20px;">
<div class="figure-box" style="margin: 0;">
<img src="src/pipe.jpg" class="figure-img" alt="Pipeline">
<p class="caption"><strong>Figure 1: Think-then-Execute.</strong> Structured planning for precise synthesis.</p>
</div>
<p style="margin-top: 15px; text-align: justify; font-size: 1rem;">
We decouple the reasoning process into a dedicated <strong>Thinker</strong> module.
This allows the model to generate structured plans before the <strong>Generator</strong>
begins the pixel-level synthesis.
</p>
</div>
</section>
<section class="section">
<h2 class="section-title">BibTeX</h2>
<div class="bibtex-box">
<pre style="margin: 0;">@misc{zhou2026unifiedthinkergeneralreasoning,
title={Unified Thinker: A General Reasoning Modular Core for Image Generation},
author={Sashuai Zhou and Qiang Zhou and Jijin Hu and Hanqing Yang and Yue Cao and Junpeng Ma and Yinchao Ma and Jun Song and Tiezheng Ge and Cheng Yu and Bo Zheng and Zhou Zhao},
year={2026},
eprint={2601.03127},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2601.03127},
}</pre>
</div>
</section>
<footer>
<p>© 2026 Sashuai Zhou. All rights reserved.</p>
</footer>
</div>
</body>
</html>