STYLE: describe OASIS dataset and vision-guided tuning in BMVC news item

Author: hyeondo-jang

index.html

@@ -431,8 +431,10 @@ <h3>Recent News</h3>
         <div class="news-item">
           <span class="news-date">Aug 2026</span>
           <span class="news-content">🎓 Our paper on semiconductor anomaly detection has been accepted to
-            BMVC 2026. In this work, we introduce a new dataset and an approach for robust anomaly
-            detection with large vision language models.</span>
+            BMVC 2026. In this work, we introduce OASIS, a large-scale SEM defect dataset for
+            instruction-tuning large vision language models, and propose a vision-guided instruction
+            tuning method that stays robust under sub-domain shifts without new human
+            annotations.</span>
         </div>
 
         <div class="news-item">

news.html

@@ -384,7 +384,7 @@
 
                 <div class="news-item">
                     <span class="news-date">Aug 2026</span>
-                    <span class="news-content">&#127891; Our paper on semiconductor anomaly detection has been accepted to BMVC 2026. In this work, we introduce a new dataset and an approach for robust anomaly detection with large vision language models.</span>
+                    <span class="news-content">&#127891; Our paper on semiconductor anomaly detection has been accepted to BMVC 2026. In this work, we introduce OASIS, a large-scale SEM defect dataset for instruction-tuning large vision language models, and propose a vision-guided instruction tuning method that stays robust under sub-domain shifts without new human annotations.</span>
                 </div>
 
                 <div class="news-item">