search

UCSC-VLAA / HQ-Edit

HQ-Edit: A High-Quality and High-Coverage Dataset for General Image Editing
https://thefllood.github.io/HQEdit_web/
Other
73 stars 3 forks source link

number of images #3

Closed betterze closed 6 months ago

betterze commented 6 months ago

Dear HQ-Edit team,

Thank you for sharing this great work.

You mention that there are 200k edits in the dataset, but I only see 99k edits after I download the dataset

from datasets import load_dataset
dataset=load_dataset('UCSC-VLAA/HQ-Edit',num_proc=16)
dataset

the output is 

DatasetDict({
    train: Dataset({
        features: ['input', 'input_image', 'edit', 'inverse_edit', 'output', 'output_image'],
        num_rows: 98675
    })
})

I try to delete the cache and redownload the dataset, and the number of images stay the same. Did I make any mistake in my command? thx alot

Best wishes, Zongze

betterze commented 6 months ago

Do you consider the reverse direction to be a different edit? so 99k*2 around 200k?

TheFllood commented 6 months ago

Yes, 99kedit + 99kinverse_edit =around 200k.

betterze commented 6 months ago

get it, thx