Update index.html

cxrl2024/index.html

@@ -195,19 +195,21 @@ <h4 class="title is-4 has-text-centered">Pipeline</h4>
           <br>
           <p>
             The pipeline of <i>CXRL</i>. 
-            Our model employs policy gradient optimization utilizing multi-reward feedbacks, fine-tuning image generator and ACE to produce realistic and accurate CXR that corresponds closely to the input report.
+            Our model employs policy gradient optimization utilizing multi-reward feedback, fine-tuning image generator, and ACE to produce realistic and accurate CXR that corresponds closely to the input report.
           </p>
         </div>
 
         <br>
-        <h4 class="title is-4 has-text-centered">Eigen Aggregation Module</h4>
+        <h4 class="title is-4 has-text-centered">Reward Feedback Models</h4>
         <div class="content has-text-justified">
           <img src="./static/images/feedback.png" alt="Main figure">
           <p>
-            An illustration of the EiCue generation. From the input image, both color affinity 
-            matrix \( \mathrm{A_{color}} \) and semantic similarity matrix \( \mathrm{A_{seg}} \) are derived, 
-            which are combined to form the Laplacian \( \mathrm{L_{sym}} \). An eigenvector 
-            subset \( \mathrm{\hat{V}} \) of \( \mathrm{L_{sym}} \) are clustered to produce EiCue.
+            A detailed illustration of our reward feedback models. We incorporate three different feedbacks for report-to-CXR generation model to generate goal-oriented CXRs.
+            <ul>
+               <li><b>Posture Alignment Feedback</b>: Generated CXRs often face scaling issues, like excessive zooming or rotation, obscuring essential details. To counter these undesirable effects, we introduce a reward signal to align the CXR's posture with a canonical orientation to preserve essential parts.</li>
+               <li><b>Diagnostic Condition Feedback</b>: To accurately reflect generated CXRs with referenced pathologies, we classify them using a parsed report label, rewarding its accuracy.</li>
+               <li><b>Multimodal Consistency Feedback</b>: We enforce the generated CXRs to better match their reports. We leverage a multimodal latent representation pretrained with CXR-report pairs for semantic agreement assessment.</li>
+             </ul>
           </p>
         </div>