Skip to content

Dataset · axxkaya

Vision-tasks dataset with textual explanations

axxkaya/UVT-Explanatory-based-Vision-Tasks

English image and text instruction dataset for training CV models to generalize to new zero-shot tasks.

What's inside

The dataset combines English images and texts, with a total size ranging from 100 thousand to 1 million examples. The authors propose replacing familiar short CV task names with detailed textual instructions and explanations. This aims to bring the generalization capabilities of computer vision models closer to modern language models.

What you can do with it

  • Training multimodal models to execute new CV tasks via text descriptions (zero-shot)
  • Unifying various computer vision tasks in a single text-visual interface
  • Research in instruction tuning for visual models
Languages
English
Size
100K–1M rows
Downloads this month
291
Likes
45

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("axxkaya/UVT-Explanatory-based-Vision-Tasks")