from transformers import pipeline from bs4 import BeautifulSoup import requests import urllib import json import argparse from pathlib import Path import os import codecs from PIL import UnidentifiedImageError from PIL import Image, ImageFile ImageFile.LOAD_TRUNCATED_IMAGES = True class Scraper(object): def __init__(self, urls): self.urls = urls def fetch(self, url): return requests.get(url, headers={"User-Agent": "Mozilla/5.0"}) def parse(self, text): return BeautifulSoup(text, 'html.parser') def download_image(self, url, path): image = requests.get(url) open(path, 'wb').write(image.content) class GraffitiDatabaseScraper(Scraper): def __init__(self): Scraper.__init__(self, ["https://graffiti-database.com"]) def get_image_urls(self, parsed): pages = parsed.select('.image-info a') image_urls = [] for page in pages: image_urls.append(page.get('href')) return image_urls def scrape_image(self, url): print("Scraping image page {0}".format(url)) fetched = self.fetch(url) if fetched.status_code != 200: raise "Scraping image page did fail" text = fetched.text parsed = self.parse(text) image = parsed.select('.img-fluid')[0].get('src') tag_links = parsed.select('a.tag') url_parts = urllib.parse.unquote( urllib.parse.unquote(url)).rsplit("/", 2) artist = url_parts[1] return { "image_url": image, "city": tag_links[-1].get_text().strip(), "artist": artist if artist != "Writer Unknown" and artist != "Writers Unknown" else None } def scrape_page(self, url): print("Scraping page {0}".format(url)) text = self.fetch(url).text parsed = self.parse(text) image_urls = self.get_image_urls(parsed) if len(image_urls) == 0: return False for image_url in image_urls: try: result = self.scrape_image( "https://graffiti-database.com" + image_url) except: continue file_name = result["image_url"].split('/')[-1] self.download_image(result["image_url"], './images/' + file_name) result["file"] = file_name with open('./images/' + file_name + '.json', 'w', encoding='utf-8') as f: f.write(json.dumps(result, indent=2, ensure_ascii=False)) return True def scrape(self): url = self.urls[0] print("Scraping {0}".format( url)) count = 1 has_images = True while has_images: has_images = self.scrape_page(url + "?page=" + str(count)) count += 1 class GraffitiScraper(Scraper): def __init__(self): Scraper.__init__(self, ["https://www.graffiti.org/index/world.html", "https://www.graffiti.org/index/europe.html", "https://www.graffiti.org/index/usa.html"]) def scrape_page(self, url, city_name): print("Scraping page {0}".format(url)) text = self.fetch(url).text parsed = self.parse(text) image_elements = parsed.select('a[href*=".jpg"]') for image_element in image_elements: if image_element.get_text().strip() == "": tags = image_element.find_next_sibling(string=True).get_text( ).strip() if image_element.find_next_sibling(string=True) else "" else: continue image_url = image_element.get("href").replace("/", "_") url_parts = url.split('/') url_parts[-1] = image_element.get("href") self.download_image("/".join(url_parts), './images/' + image_url) with open('./images/' + image_url + '.json', 'w', encoding='utf-8') as f: f.write(json.dumps({ "file": image_url, "image_url": "/".join(url_parts), "artist": tags, "city": city_name if city_name != "Various cities" else None }, indent=2, ensure_ascii=False)) def scrape_url(self, url): print("Scraping url {0}".format(url)) text = self.fetch(url).text parsed = self.parse(text) cities = parsed.find_all("h4") for city in cities: city_name = city.get_text().split("\n")[0].strip() pages = city.find_all("a") for page in pages: if page.get_text().strip() == "ยง": continue self.scrape_page( "https://www.graffiti.org/index/" + page.get("href"), city_name) def scrape(self): for url in self.urls: self.scrape_url(url) class CLI(): def __init__(self): parser = argparse.ArgumentParser( prog='graffiti-cli', description='Tools for setting up the dataset') subparsers = parser.add_subparsers(dest="command", required=True) scrape = subparsers.add_parser( 'scrape', help='Scrapes data sources and downloads images') scrape.add_argument('--source', default='graffiti.org', choices=['graffiti.org', 'graffiti-database.com'], help='Choose data source to scrape') subparsers.add_parser('cleanup', help='Cleans up downloaded images') subparsers.add_parser('caption', help='Captions downloaded images') metadata = subparsers.add_parser('metadata', help='Creates single meta files from metadata.jsonl') metadata.add_argument('--source', default='graffiti.org', choices=['graffiti.org', 'graffiti-database.com'], help='Choose data source to use') args = parser.parse_args() if args.command == 'scrape': if args.source == 'graffiti.org': GraffitiScraper().scrape() elif args.source == 'graffiti-database.com': GraffitiDatabaseScraper().scrape() elif args.command == 'cleanup': path = Path("./images").rglob("*.jpg") for i, img_p in enumerate(path): try: Image.open(img_p) except UnidentifiedImageError: path_name = str(img_p) print(path_name + " is broken. Deleting!") os.remove(path_name) os.remove(path_name + ".json") elif args.command == 'caption': captioner = pipeline( "image-to-text", model="Salesforce/blip-image-captioning-base") path = Path("./images").rglob("*.jpg") count = len(list(Path("./images").rglob("*.jpg"))) for i, img_p in enumerate(path): path_name = str(img_p) with open(path_name + ".json", 'r+', encoding='utf-8') as f: data = json.load(f) f.seek(0) caption = captioner(path_name)[0]["generated_text"] if "album" in caption: caption = "a wall with graffiti on it" json.dump({ "file": data["file"], "image_url": data["image_url"], "artist": data["artist"], "city": data["city"], "caption": caption }, f, indent=2) f.truncate() print("{0} / {1}".format(i + 1, count), path_name, caption) elif args.command == 'metadata': with open("data/" + args.source + "/metadata.jsonl", encoding="utf-8") as f: for row in f: data = json.loads(row) with open('./images/' + data["file"] + '.json', 'w') as j: j.write(json.dumps(data, indent=2, ensure_ascii=False)) def main(): CLI() if __name__ == "__main__": main()