change font

RLHF-V · Nov 30, 2023 · f1874a2 · f1874a2
1 parent fb091e9
commit f1874a2
Show file tree

Hide file tree

Showing 20 changed files with 7 additions and 7 deletions.
diff --git a/.DS_Store b/.DS_Store
diff --git a/demos/p1.png b/demos/p1.png
diff --git a/demos/p2.png b/demos/p2.png
diff --git a/demos/p3.png b/demos/p3.png
diff --git a/demos/p4.png b/demos/p4.png
diff --git a/demos/p5.png b/demos/p5.png
diff --git a/demos/p6.png b/demos/p6.png
diff --git a/demos/p7.png b/demos/p7.png
diff --git a/demos/p8.png b/demos/p8.png
diff --git a/demos/p9.png b/demos/p9.png
diff --git a/images/.DS_Store b/images/.DS_Store
diff --git a/images/Intro.png b/images/Intro.png
diff --git a/images/VPGTrans.png b/images/VPGTrans.png
diff --git a/images/blip2.png b/images/blip2.png
diff --git a/images/cost.png b/images/cost.png
diff --git a/images/icon.jpg b/images/icon.jpg
diff --git a/images/icon.png b/images/icon.png
diff --git a/images/overview.png b/images/overview.png
diff --git a/images/vl-llama.png b/images/vl-llama.png
diff --git a/index.html b/index.html
@@ -33,10 +33,10 @@
   <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="icon" href="images/icon.png">
+  <link rel="icon" href="images/icon.jpg">
   <link rel="stylesheet" href="./static/css/index.css">
 
-  <link rel="shortcut icon" href="images/icon.png" type="image/x-icon">
+  <link rel="shortcut icon" href="images/icon.jpg" type="image/x-icon">
 
   <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
   <script defer src="./static/js/fontawesome.all.min.js"></script>
@@ -82,7 +82,7 @@
       <div class="columns is-centered">
         <div class="column has-text-centered">
           <h1 class="title is-1 publication-title">RLHF-V</h1>
-          <h2 class="title is-2 publication-title">Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedback</h2>
+          <h2 class="title is-3 publication-title">Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedback</h2>
           <div class="is-size-5">
             <span class="author-block">
                 <a href="https://github.com/yiranyyu" style="color:#008AD7;font-weight:normal;">Tianyu Yu<sup>1</sup>
@@ -359,16 +359,16 @@ <h2 class="title is-3">Abstract</h2>
             Existing Multimodal Large Language Models prevalently suffer from serious <b>hallucination</b> problems, generating text that is not factually grounded in associated images. Our <b>RLHF-V framework</b> enhances MLLM trustworthiness via behavior alignment from fine-grained correctional human feedback.
             <ul>
               <li>
-                <b>1.4K Fine-Grained and Diverse Human Preference Data</b>: We collect 1.4K pieces of segment-level corrections of human feedback on hallucinations, covering hallucination types including objects (41.2%), positions (20.3%), numbers (16.5%), attributes (10%), actions (5.3%), and others (6.8%).
+                <b>1.4K Fine-Grained and Diverse Human Preference Data</b>: <span style="font-size: 95%;">We collect 1.4K pieces of segment-level corrections of human feedback on hallucinations, covering hallucination types including objects (41.2%), positions (20.3%), numbers (16.5%), attributes (10%), actions (5.3%), and others (6.8%).</span>
               </li>
               <li>
-                <b>High Data Efficiency and Scalability</b>: With just 1.4K annotated data, we achieved a 34.8% reduction in model hallucinations. Moreover, the decrease in hallucinations becomes more significant as more data used.
+                <b>High Data Efficiency and Scalability</b>: <span style="font-size: 95%;">With just 1.4K annotated data, we achieved a 34.8% reduction in model hallucinations. Moreover, the decrease in hallucinations becomes more significant as more data used.</span>
               </li>
               <li>
-                <b>Enhanced Performance and Computational Efficiency with DDPO</b>: Our Dense Direct Preference Optimization (DDPO) algorithm can better exploit the fine-grained human feedback, allowing training in under 1 hour on 8 A100 GPUs.
+                <b>Enhanced Performance and Computational Efficiency with DDPO</b>: <span style="font-size: 95%;">Our Dense Direct Preference Optimization (DDPO) algorithm can better exploit the fine-grained human feedback, allowing training in under 1 hour on 8 A100 GPUs.</span>
               </li>
               <li>
-                <b>Outstanding Trustworthiness without Compromising Helpfulness</b>: Our model surpasses existing open-source MLLMs in reducing hallucination rates, mitigates hallucination from over-generalization, and maintains informativeness.
+                <b>Outstanding Trustworthiness without Compromising Helpfulness</b>: <span style="font-size: 95%;">Our model surpasses existing open-source MLLMs in reducing hallucination rates, mitigates hallucination from over-generalization, and maintains informativeness.</span>
               </li>
             </ul>
 <!--             <br>