Skip to content

Dataset · helenqu

ImageNet images with pasted extraneous objects

helenqu/ImageNet-Paste

10 to 100 thousand images to test model robustness against visual noise.

What's inside

The dataset is collected based on the ImageNet validation split: a small picture of another concept is pasted into each original image. The dataset size ranges from 10K to 100K examples in English, with the file format unspecified.

What you can do with it

  • Evaluating the robustness of classifiers and multimodal models to visual distractions
  • Testing the ability of neural networks to isolate the main object in complex scenes
Languages
English
Size
10K–100K rows
Downloads this month
165K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("helenqu/ImageNet-Paste")