Skip to content

Dataset · songweig

Black-and-white sketches for 1000 ImageNet classes

songweig/imagenet_sketch

50,000 black-and-white sketches, 50 per ImageNet class, for evaluating computer vision robustness.

What's inside

The dataset contains 50,000 images with English label domains, exactly 50 black-and-white sketches for each of the 1000 ImageNet classes. Images were gathered via Google Images search for "sketch of ...", manually cleaned from irrelevant examples, and missing samples up to 50 per class were augmented via flips and rotations.

What you can do with it

  • Testing image classification models for domain generalization robustness
  • Fine-tuning vision models and adapters for sketch and line drawing recognition
Tasks
Image classification
Languages
English
Size
10K–100K rows
Downloads this month
90K
Likes
24

How to load it

from datasets import load_dataset

ds = load_dataset("songweig/imagenet_sketch")