Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -35,7 +35,10 @@ size_categories:
|
|
35 |
|
36 |
## Dataset Summary
|
37 |
|
38 |
-
|
|
|
|
|
|
|
39 |
|
40 |
|
41 |
### Classifiers Performance
|
|
|
35 |
|
36 |
## Dataset Summary
|
37 |
|
38 |
+
|
39 |
+
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.
|
40 |
+
|
41 |
+
|
42 |
|
43 |
|
44 |
### Classifiers Performance
|