collectDALLE3fromLAION.py

from datasets import load_dataset
import requests
from PIL import Image
from tqdm import tqdm

dataset = load_dataset("laion/dalle-3-dataset",split="train")


for i,row in enumerate(tqdm(dataset)):
    with open(f"dalle3/{i:06}.txt","w") as f:
        f.write(row["caption"])

    img_url = row["link"]
    raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')
    raw_image.save(f"dalle3/{i:06}.png")

If you want to save the images with more compression, change the code:

- raw_image.save(f"dalle3/{i:06}.png")
+ raw_image.save(f"dalle3/{i:06}.webp", quality=90)