Mobius/index.html at main · Agentic-CliniAI/Mobius · GitHub

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
<!DOCTYPE html>
<html lang="en">
<head>
  <meta charset="UTF-8" />
  <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
  <title>ℳöbius Framework — Medical Multimodal Mutual Supervision</title>
  <style>
    :root {
      --primary: #1a73e8;
      --light-bg: #f8fafc;
      --text: #202124;
    }
    * {
      margin: 0;
      padding: 0;
      box-sizing: border-box;
    }
    body {
      font-family: 'Segoe UI', system-ui, sans-serif;
      line-height: 1.6;
      color: var(--text);
      background-color: var(--light-bg);
    }
    header {
      background: linear-gradient(135deg, #0d47a1, #1a73e8);
      color: white;
      padding: 3rem 2rem;
      text-align: center;
    }
    h1 {
      font-size: 2.4rem;
      margin-bottom: 0.5rem;
    }
    .subtitle {
      font-size: 1.2rem;
      opacity: 0.9;
      max-width: 800px;
      margin: 0 auto 1.2rem;
    }
    .header-icons {
      display: flex;
      justify-content: center;
      gap: 24px;
      margin-top: 8px;
    }
    .header-icons a {
      display: inline-flex;
      align-items: center;
      justify-content: center;
      width: 40px;
      height: 40px;
      border-radius: 50%;
      background: rgba(255, 255, 255, 0.1);
      color: white;
      text-decoration: none;
      transition: background 0.2s, transform 0.2s;
    }
    .header-icons a:hover {
      background: rgba(255, 255, 255, 0.2);
      transform: scale(1.1);
    }
    .header-icons img {
      width: 32px;
      height: 32px;
      display: block;
    }
    main {
      max-width: 900px;
      margin: 2rem auto;
      padding: 0 1.5rem;
    }
    section {
      margin-bottom: 2.5rem;
    }
    h2 {
      color: var(--primary);
      margin-bottom: 1rem;
      font-size: 1.6rem;
    }
    p {
      margin-bottom: 1rem;
    }
    ul {
      padding-left: 1.5rem;
      margin: 1rem 0;
    }
    li {
      margin-bottom: 0.4rem;
    }
    .highlight {
      background-color: #e3f2fd;
      padding: 0.1rem 0.3rem;
      border-radius: 3px;
      font-weight: 500;
    }
    footer {
      text-align: center;
      padding: 2rem;
      color: #5f6368;
      font-size: 0.9rem;
    }
    footer a {
      color: var(--primary);
      text-decoration: none;
    }
  </style>
</head>
<body>
  <header>
    <h1>ℳöbius Framework</h1>
    <p class="subtitle">Closed-Loop Mutual Supervision for Reliable Medical Vision–Language Reasoning</p>
    <div class="header-icons">
      <a href="https://github.com/NJU-MedAI-Lab/mobius" target="_blank" rel="noopener" aria-label="GitHub">
        <img src="./icons/github.svg" alt="GitHub" />
      </a>
      <a href="https://huggingface.co/NJU-MedAI-Lab" target="_blank" rel="noopener" aria-label="Hugging Face">
        <img src="./icons/huggingface-color.svg" alt="Hugging Face" />
      </a>
    </div>
  </header>

  <main>
    <section>
      <h2>Overview</h2>
      <p>
        The <span class="highlight">Möbius Framework</span> addresses <strong>hallucination</strong> and <strong>modality laziness</strong> in medical large vision-language models (LVLMs) by establishing a bidirectional, closed-loop supervision between an LVLM and a Segment Anything Model (SAM).
      </p>
      <p>
        Language from the LVLM guides anatomical segmentation, while segmentation from SAM grounds visual reasoning—enabling annotation-efficient alignment of perception and semantics.
      </p>
    </section>

    <section>
      <h2>Core Components</h2>
      <div class="models">
        <div class="model-card">
          <div class="model-name">ℳöbiusMed-LVLM</div>
          <p>A medical LVLM trained within the Möbius framework, producing visually grounded diagnostic descriptions with reduced hallucination.</p>
        </div>
        <div class="model-card">
          <div class="model-name">ℳöbiusMed-SAM</div>
          <p>A vision foundation model enhanced by linguistic guidance from the LVLM, capable of interpreting free-form medical instructions.</p>
        </div>
      </div>
    </section>

    <section>
      <h2>Key Innovation</h2>
      <p>
        Unlike conventional pipelines that treat vision and language modules independently, Möbius creates a <span class="highlight">self-correcting loop</span>:
      </p>
      <ul style="padding-left: 1.5rem; margin: 1rem 0;">
        <li>LVLM → SAM: Natural language prompts refine region-of-interest segmentation.</li>
        <li>SAM → LVLM: Precise masks constrain visual grounding, suppressing unfaithful generation.</li>
      </ul>
      <p>
        This mutual supervision mimics the Möbius strip—seamlessly connecting two modalities into a unified, self-reinforcing system.
      </p>
    </section>
  </main>

  <footer>
    <p>Proposed by Xiao Song and Haonan Qin, <a href="https://prlab-nju.com/">PR-Lab</a> @ Nanjing University • 2025</p>
    <p>For research purposes only. Part of doctoral work on medical multimodal reliability.</p>
  </footer>
</body>
</html>