Dataset Viewer
Auto-converted to Parquet Duplicate
image
imagewidth (px)
640
5.76k
prompt
stringlengths
16
82
referral_type
stringclasses
3 values
editing_task
stringclasses
4 values
Replace the black seeds with sunflower seeds.
Property-based
Replacement
Change the color of the red shoes to blue.
Property-based
Attribute Change
Replace the red chili peppers with green jalapenos.
Property-based
Replacement
Change the ring to be made of gold.
Property-based
Attribute Change
Change the color of the grey tank top to blue.
Property-based
Attribute Change
Change Sailor Jupiter's uniform to yellow.
Knowledge-based
Attribute Change
Change the skirt of Chibi Maruko-chan to purple.
Knowledge-based
Attribute Change
Make Thanos smile.
Knowledge-based
Attribute Change
Replace the ceiling light with a modern crystal chandelier.
Property-based
Replacement
Make the happy face on the oatmeal look sad.
Knowledge-based
Attribute Change
Make the grumpy-looking monkey smile.
Knowledge-based
Attribute Change
Change the color of the white car to red.
Property-based
Attribute Change
Change Killua's shirt to blue.
Knowledge-based
Attribute Change
Change the pink frosting to blue.
Property-based
Attribute Change
Change the color of the orange drink to blue.
Property-based
Attribute Change
Change the color of the red dress to blue.
Property-based
Attribute Change
Change the wooden board to a white marble slab.
Property-based
Attribute Change
Make the lens with the feather pattern solid red.
Property-based
Attribute Change
Change the color of the red rose to white.
Property-based
Attribute Change
Replace Inosuke's boar mask with his human face.
Knowledge-based
Replacement
Remove the pet carrier.
Property-based
Removal
Make the dog look happier.
Knowledge-based
Attribute Change
Change the color of the grey t-shirt to blue.
Property-based
Attribute Change
Change the father's shirt to blue.
Knowledge-based
Attribute Change
Change the black and white dog into a brown dog.
Property-based
Attribute Change
Change the color of the car to red.
Property-based
Attribute Change
Change the father's shirt to a light blue color.
Knowledge-based
Attribute Change
Replace the dragon fruit with a pineapple.
Property-based
Replacement
Change the color of the drink to pink.
Property-based
Attribute Change
Change the color of the yellow measuring tape to blue.
Property-based
Attribute Change
Remove the pearl necklace.
Property-based
Removal
Change the color of the reddish-orange feathers to blue.
Property-based
Attribute Change
Change the black leather on the chairs to white.
Property-based
Attribute Change
Remove the black chopsticks.
Property-based
Removal
Replace the french fries with onion rings.
Property-based
Replacement
Change the color of the yellow sofa to a deep blue.
Property-based
Attribute Change
Change the material of the vase to glass.
Property-based
Attribute Change
Change the yellow shoe to a green color.
Property-based
Attribute Change
Change the color of the fluffy white dog to black.
Property-based
Attribute Change
Change the color of the red dress to green.
Property-based
Attribute Change
Remove the rosemary sprig.
Property-based
Removal
Change the color of the pear-shaped sweet to light green.
Property-based
Attribute Change
Remove the white dipping sauce.
Property-based
Removal
Make the paintbrush look clean and new.
Property-based
Attribute Change
Make the mother look happier.
Knowledge-based
Attribute Change
Make Tanjiro look happy.
Knowledge-based
Attribute Change
Change the white dress to blue.
Property-based
Attribute Change
Change the color of the pink sofa to a dark gray.
Property-based
Attribute Change
Change the color of the red hat to blue.
Property-based
Attribute Change
Change the color of the yellow towel to blue.
Property-based
Attribute Change
Make Master Shifu look happy.
Knowledge-based
Attribute Change
Remove the dog on the left.
Spatial-based
Removal
Change the color of the pink pants to dark gray.
Property-based
Attribute Change
Make the scared man look happy.
Knowledge-based
Attribute Change
Change Dilraba Dilmurat's hair to platinum blonde.
Knowledge-based
Attribute Change
Change the pattern of the polka-dot plate to solid blue.
Property-based
Attribute Change
Replace the potato in the top right with a mushroom.
Spatial-based
Replacement
Make Shin-chan look happy.
Knowledge-based
Attribute Change
Change the color of the youngest person's sweater to blue.
Knowledge-based
Attribute Change
Replace the bamboo muzzle with a red rose.
Property-based
Replacement
Change the color of the white flowers to pink.
Property-based
Attribute Change
Change the white macaron to strawberry pink.
Property-based
Attribute Change
Change the black soap bar to blue.
Property-based
Attribute Change
Change the purple tracksuit into a blue one.
Property-based
Attribute Change
Change the color of the gray mug to red.
Property-based
Attribute Change
Remove the runner on the far right.
Spatial-based
Removal
Change the color of the orange chair to dark blue.
Property-based
Attribute Change
Change the hair of the person with red hair to purple.
Property-based
Attribute Change
Change the color of the yellow filling to chocolate brown.
Property-based
Attribute Change
Change the material of the leather chaise lounge to velvet.
Property-based
Attribute Change
Replace the shrimp with slices of strawberry.
Property-based
Replacement
Change the color of the rug to gray.
Property-based
Attribute Change
Change the dark green teapot to a blue and white porcelain style.
Property-based
Attribute Change
Remove the dog in the middle.
Spatial-based
Removal
Change the color of the brown owl to pink.
Property-based
Attribute Change
Change the color of the white shirt to red.
Property-based
Attribute Change
Change the color of the pink macaron to yellow.
Property-based
Attribute Change
Change the harness on the dog on the left to have a plaid pattern.
Spatial-based
Attribute Change
Change the car on the left to a silver color.
Spatial-based
Attribute Change
Change the color of the red tail feathers to blue.
Property-based
Attribute Change
Replace the plums with lemons.
Property-based
Replacement
Change the color of the dark green armchairs to camel brown.
Property-based
Attribute Change
Change the color of the purple yarn to teal.
Property-based
Attribute Change
Change the chocolate sauce to strawberry sauce.
Property-based
Attribute Change
Change the grey cat to an orange cat.
Property-based
Attribute Change
Change the yellow lamp to a red color.
Property-based
Attribute Change
Remove the dumpling on the far left.
Spatial-based
Removal
Change the purple sweater on the dog to a denim jacket.
Property-based
Attribute Change
Change Elsa's hair to brown.
Knowledge-based
Attribute Change
Change the material of the wooden vase to black ceramic.
Property-based
Attribute Change
Remove the Adidas logo from the sweater.
Knowledge-based
Removal
Change the red lipstick to a vibrant pink.
Property-based
Attribute Change
Remove the koala on the right.
Spatial-based
Removal
Change the color of the blue hat to purple.
Property-based
Attribute Change
Remove the peeled orange segment.
Property-based
Removal
Change the brown leather sofa to a dark green color.
Property-based
Attribute Change
Change the color of the grey socks to blue.
Property-based
Attribute Change
Change the airborne bird into a grey heron.
Knowledge-based
Replacement
Remove the panda on the far right.
Spatial-based
Removal
Change the color of the blue and purple bottle to a solid gold.
Property-based
Attribute Change
End of preview. Expand in Data Studio

SREdit-Bench: A Benchmark for Spatial Reasoning in Image Editing

Project Page Paper GitHub

Dataset Description

SREdit-Bench is a challenging benchmark designed to comprehensively evaluate the spatial reasoning capabilities of image editing models. While existing benchmarks primarily focus on object-salient scenes, they often under-represent spatially complex editing scenarios (e.g., multiple similar editable entities, non-object-salient scenes, and fine-grained object referral).

To fill this gap, we introduce SREdit-Bench, comprising 590 carefully curated high-resolution (>1024×1024) samples. It focuses on evaluating a model's true spatial reasoning ability under complex editing scenarios

Key Features

  1. Sophisticated Scenes: The benchmark deliberately includes images with multiple entities and non-object-centric layouts to test robust localization.
  2. Fine-grained Target Referral: The instructions require deep spatial and contextual understanding, categorized into three modes:
    • Spatial: Instructions with explicit location or relational cues.
    • Property: Instructions using appearance or attribute-based descriptions.
    • Knowledge: Implicit, context-dependent cues that require background or commonsense knowledge.

Dataset Structure

(Note: Please adjust the field names based on your actual Hugging Face dataset columns)

Each instance in the dataset typically contains:

  • image: The source high-resolution image.
  • instruction: The editing instruction containing fine-grained referring expressions.
  • referral_type: The category of the instruction (Spatial, Property, or Knowledge).
  • edit_type: The category of the editing task (e.g., Add, Remove, Replace, Modify).

Evaluation Protocol

We follow the VIEScore, we evaluate the edited images across three main metrics:

  1. Semantic Consistency (SC): Measures how well the edited result faithfully follows the given instruction.
  2. Perceptual Quality (PQ): Evaluates the image's naturalness and the presence of visual artifacts.
  3. Overall (O): The geometric mean of SC and PQ, averaged across all test samples.

Usage

You can easily load this dataset using the datasets library:

from datasets import load_dataset

dataset = load_dataset("zjYinnnn/SREdit-Bench")
print(dataset['test'][0])
Downloads last month
-

Paper for zjYinnnn/SREdit-Bench