Update README.md

README.md

@@ -35,7 +35,10 @@ size_categories:
 
 ## Dataset Summary
 
-The **ImageNet-Hard** contains challenging images sourced from the validation datasets of ImageNet.
+
+The **ImageNet-Hard** is a new benchmark that comprises an array of challenging images, curated from several validation datasets of ImageNet. This dataset challenges state-of-the-art vision models, as merely zooming in often fails to enhance their ability to correctly classify images. Consequently, even the most advanced models, such as `CLIP-ViT-L/14@336px`, struggle to perform well on this dataset, achieving only `2.02%` accuracy.
+
+
 
 
 ### Classifiers Performance