img_id
stringlengths
2
6
turn_index
int32
1
3
source_img
imagewidth (px)
500
1.02k
mask_img
imagewidth (px)
1.02k
1.02k
instruction
stringlengths
8
72
target_img
imagewidth (px)
1.02k
1.02k
327726
1
change the table for a dog
435242
1
leave only words on the page
435242
2
leave words only on half the page
513551
1
leave only the scissor on the cup
138270
1
the window is now square
181992
1
let the car be a sports car
181992
2
remove the mesh screen
181992
3
replace some bags with throw pillows
442654
1
turn the stop sign into a no parking sign
442654
2
remove the lower sign
403221
1
Let the sheep turn white.
278535
1
Have the business be called Bongos.
278535
2
Turn the stop sign into a lollipop.
298692
1
make that backpack all black
298692
2
make the girl ginger
262146
1
Give the person a cape.
275542
1
let the dogs bite a stuffed toy
275542
2
let the dogs wear jackets
503855
1
Get rid of all the people.
503855
2
Make the umbrellas blue.
503855
3
Add a big dog.
166761
1
change the grass to water
166761
2
make the airplane a toy airplane
173530
1
What if it was another car on it?
504170
1
The background should be of a mountain.
43894
1
let the cellphone screen be blue
43894
2
let the laptop be black
10909
1
remove the microwave and stove and put a refrigerator in it's place
10909
2
swap the cabinet for a shelf
236951
1
add a bullet train
236951
2
remove the boards
236951
3
add a bear
61184
1
remove the bench with pillows
61184
2
remove one potted plant
507950
1
let the train be blue
507950
2
let a lion walk towards the cows
507950
3
add a helicopter in the sky
566823
1
Let the bus be single-sectioned.
442416
1
add a phone
442416
2
let the two pizzas be pepperoni pizzas
442416
3
add a microwave
357633
1
put a car on the screen of the laptop.
357633
2
It could be a hand next to the laptop.
357633
3
The hand could be holding a cup.
115859
1
make one of the dogs pant
383706
1
What if the doughnut had bites?
570259
1
the bowl needs to have ketchup.
570259
2
Can it be a hotdog insted of sandwich?
570259
3
Put a pizza next to the hotdog.
74549
1
Let the sandwich not be eaten.
23781
1
add a bouquet of red flowers
23781
2
change potatoes into oranges
23781
3
let a grandmother look at the display
192047
1
Replace the mirror with a window.
192047
2
Put a towel on the hanger.
152327
1
Make it one-legged.
251888
1
let the boat be white boat
251888
2
add a shark in the water
251888
3
add a plane in the sky
528432
1
let it be steel bats
528432
2
add a baseball on the glass table
528432
3
add a rat
328270
1
change the blue shirt to red shirt
328270
2
let it be a mushroom pizza
328270
3
add a bottle of olive oil
553990
1
Make the horse's coat dark black.
553990
2
Make the woman's pants black.
171472
1
let the bag contain soda bottles
171472
2
let there be a couch near the bags
171472
3
let a dog coil up and sleep
551427
1
make the stove be on fire
546597
1
make the street empty
546597
2
edit a castle in the image instead of the buildings
540122
1
turn the computer off
459662
1
Have one of the girls be wearing a baseball cap
62756
1
make his tie black
62756
2
replace the white truck with a white motorcycle
62756
3
give the man bermuda shorts
292512
1
close the window
533536
1
Make the cat a dog.
413182
1
Have both kids be wearing red baseball caps
413182
2
Have there be the number 9 on the shirt of one of the kids
357057
1
remove the towels
357057
2
remove that toilet paper
357057
3
remove the items from every shelf
50058
1
let the snow pile be a grassy hill
50058
2
let the man wear a sombrero
50058
3
let the man wear casual clothes
274516
1
change the motorcycle into an ATV
274516
2
let the man wear a red jacket
274516
3
make a rabbit chase the man
549042
1
make the elephant raise its trunk
549042
2
let the elephants have bigger eyes
549042
3
let the elephant wear a beret
114748
1
Let's add a flower sticker on the fridge.
373038
1
make the cat lick his paw
17324
1
let the zebra have its legs up in the air
17324
2
add birds flying in the sky
17324
3
let the tall brown grass be green
72795
1
Replace the batter with Batman.

Dataset Card for MagicBrush

Dataset Summary

MagicBrush is the first large-scale, manually-annotated instruction-guided image editing dataset covering diverse scenarios single-turn, multi-turn, mask-provided, and mask-free editing. MagicBrush comprises 10K (source image, instruction, target image) triples, which is sufficient to train large-scale image editing models.

Please check our website to explore more visual results.

Dataset Structure

"img_id" (str): same from COCO id but in string type, for easier test set loading

"turn_index" (int32): the edit turn in the image

"source_img" (str): input image, could be the original real image (turn_index=1) and edited images from last turn (turn_index >=2)

"mask_img" (str): free-form mask image (white region), can be used in mask-provided setting to limit the region to be edited.

"instruction" (str): edit instruction of how the input image should be changed.

"target_img" (str): the edited image corresponding to the input image and instruction.

If you need auxiliary data, please use training set and dev set

Splits

train: 8,807 edit turns (4,512 edit sessions). dev: 528 edit turns (266 edit sessions). test: (To prevent potential data leakage, please check our repo for information on obtaining the test set.)

Licensing Information

Creative Commons License This work is licensed under a Creative Commons Attribution 4.0 International License.

Citation Information

If you find this dataset useful, please consider citing our paper:

@inproceedings{Zhang2023MagicBrush,
        title={MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing},
        author={Kai Zhang and Lingbo Mo and Wenhu Chen and Huan Sun and Yu Su},
        booktitle={Advances in Neural Information Processing Systems},
        year={2023}
}
Downloads last month
2,965

Models trained or fine-tuned on osunlp/MagicBrush