Skip to content

Dataset · facebook

Benchmark for evaluating image editing

facebook/emu_edit_test_set

A test set of image-text pairs for evaluating image editing models across 7 categories.

What's inside

A test dataset from Facebook containing 1 to 10k examples, including images and text descriptions. Created as a benchmark based on images from MagicBrush, it covers seven editing types: background replacement, global changes, style change, object removal and addition, local edits, and color or texture modification.

What you can do with it

  • Evaluating the quality of text-guided image editing models
  • Comparing local and global editing algorithms on a standard benchmark
Size
1K–10K rows
Downloads this month
422
Likes
48

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/emu_edit_test_set")