Update README.md

README.md

@@ -38,10 +38,6 @@ for Comprehensive Video Understanding". It mainly consists of three parts: `vide
 
 - `subtitle.zip` contains **800 JSONL files**, named `{001..800}.jsonl`, each corresponding to a video id. Each file contains word-level entries with timestamps.
 
-<p align="center">
-    <img src="assets/teaser.png" width="100%" height="100%">
-</p>
-
 ---
 
 # 🩷 About This Benchmark
@@ -75,18 +71,22 @@ In 2024, our [**Video-MME**](https://video-mme.github.io/) benchmark became a st
 
   A non-linear scoring mechanism is applied to all question groups, and a first error truncation mechanism is used for reasoning coherence groups.
 
+<p align="center">
+    <img src="assets/teaser.png" width="100%" height="100%">
+</p>
+
 ---
 
 # 🍺 Example
 
-<p align="center">
+<p align="left">
   <a href="https://huggingface.co/datasets/MME-Benchmarks/Video-MME-v2/resolve/main/assets/demo.mp4">
-    <img src="assets/demo_cover.png" width="60%" alt="Demo video"/>
+    <img src="assets/demo_cover.png" width="60%" alt="Demo video cover"/>
   </a>
 </p>
 
-<p align="center">
-  <strong>▶ Click the image above to open the demo video.</strong>
+<p align="left">
+  <strong>👆 Click the cover image to view the demo video.</strong>
 </p>
 
 <p>