Skip to content

Dataset · gmongaras

ImageNet-21K with updated text descriptions

gmongaras/Imagenet21K

Around 13 million PNG images across ~19,000 classes with new author-generated captions.

What's inside

The full version of the large ImageNet-21K dataset, containing approximately 13 million examples and around 19,000 text classes. Images are saved in PNG format as raw bytes, with newly generated captions for all pictures by the author.

What you can do with it

  • Pre-training large computer vision models on thousands of visual categories
  • Training multimodal models to associate images with text labels
Size
10M–100M rows
Downloads this month
6.3K
Likes
13

How to load it

from datasets import load_dataset

ds = load_dataset("gmongaras/Imagenet21K")