icrawler

Introduction

This package is a mini framework of image crawlers. Scrapy is heavy and powerful, while icrawler is tiny and flexible.

With icrawler, you can write a crawler easily by focusing on the contents you want to crawl, avoiding some troublesome problems like exception handling, thread scheduling and communication. It also provides built-in crawlers for popular image sites such as search engines (Google, Bing, Baidu) and flickr.

Requirements

Python 2.7+ or 3.4+.

Stucture

It consists of 3 main components (Feeder, Parser and Downloader) and 2 FIFO queues (url_queue and task_queue). The workflow is shown in the following figure.

url_queue stores the url of pages which may contain images
task_queue stores the image url as well as any meta data you like, each element in the queue is a dictionary and must contain the field img_url
Feeder puts page urls to url_queue
Parser requests and parses the page, then extracts the image urls and puts them into task_queue
Downloader gets tasks from task_queue and requests the images, then saves them in the given path.

Feeder, parser and downloader are all thread managers, which means they start threads to finish corresponding tasks, so you can specify the number of threads they use.

Quick start

Installation

The quick way:

pip install icrawler

You can also manually install it by

git clone git@github.com:hellock/icrawler.git
cd icrawler
python setup.py install

If you fail to install icrawler on Linux, it is probably caused by lxml. See here for solutions.

Use built-in crawlers

This framework contains 5 built-in crawlers.

Google
Bing
Baidu
Flickr
General greedy crawl (crawl all the images from a website)

Here is an example of how to use the built-in crawlers. The search engine crawlers have similar interfaces.

from icrawler.examples import GoogleImageCrawler
from icrawler.examples import BingImageCrawler
from icrawler.examples import BaiduImageCrawler

google_crawler = GoogleImageCrawler('your_image_dir')
google_crawler.crawl(keyword='sunny', offset=0, max_num=1000,
                     date_min=None, date_max=None, feeder_thr_num=1,
                     parser_thr_num=1, downloader_thr_num=4,
                     min_size=(200,200), max_size=None)
bing_crawler = BingImageCrawler('your_image_dir')
bing_crawler.crawl(keyword='sunny', offset=0, max_num=1000,
                   feeder_thr_num=1, parser_thr_num=1, downloader_thr_num=4,
                   min_size=None, max_size=None)
baidu_crawler = BaiduImageCrawler('your_image_dir')
baidu_crawler.crawl(keyword='sunny', offset=0, max_num=1000,
                    feeder_thr_num=1, parser_thr_num=1, downloader_thr_num=4,
                    min_size=None, max_size=None)

Note: Only google image crawler supports date range parameters.

Flickr crawler is a little different.

from datetime import date
from icrawler.examples import FlickrImageCrawler

flickr_crawler = FlickrImageCrawler('your_apikey', 'your_image_dir')
flickr_crawler.crawl(max_num=1000, feeder_thr_num=1, parser_thr_num=1,
                     downloader_thr_num=1, tags='child,baby',
                     group_id='68012010@N00', min_upload_date=date(2015, 5, 1))

Supported optional searching auguments are

user_id -- The NSID of the user who's photo to search.
tags -- A comma-delimited list of tags.
tag_mode -- Either 'any' for an OR combination of tags, or 'all' for an AND combination.
text -- A free text search. Photos who's title, description or tags contain the text will be returned.
min_upload_date -- Minimum upload date. The date can be in the form of datetime.date object, a unix timestamp or a string.
max_upload_date -- Maximum upload date. Same form as min_upload_date.
group_id -- The id of a group who's pool to search.
extras -- A comma-delimited list of extra information to fetch for each returned record. See here for more details.
per_page -- Number of photos to return per page.

If you just want to crawl all the images from some website, then GreedyImageCrawler may be helpful.

from icrawler.examples import GreedyImageCrawler

greedy_crawler = GreedyImageCrawler('images/greedy')
greedy_crawler.crawl(domains='bbc.com', max_num=0,
                     parser_thr_num=1, downloader_thr_num=1,
                     min_size=None, max_size=None)

The argument domains can be either a url string or list. Second level domains and subpaths are supported, but there should be no scheme like 'http' in the domains.

You can see the complete example in test.py, to run it

python test.py [options]

options can be google, bing , baidu, flickr, greedy or all, using all by default if no auguments are specified. Note that you have to provide your flickr apikey if you want to test FlickrCrawler.

Write your own crawler

The simplest way is to override some methods of Feeder, Parser and Downloader class.

Feeder

The method you need to override is

feeder.feed(**kwargs)

If you want to offer the start urls at one time, for example from 'http://example.com/page_url/1' up to 'http://example.com/page_url/10'

from icrawler import Feeder

class MyFeeder(Feeder):
    def feed(self):
        for i in range(10):
            url = 'http://example.com/page_url/{}'.format(i + 1)
            self.url_queue.put(url)

Parser

The method you need to override is

parser.parse(response, **kwargs)

response is the page content of the url from url_queue, what you need to do is to parse the page and extract image urls, and then put them into task_queue. Beautiful Soup package is recommended for parsing html pages. Taking GoogleParser for example,

class GoogleParser(Parser):

    def parse(self, response):
        soup = BeautifulSoup(response.content, 'lxml')
        image_divs = soup.find_all('div', class_='rg_di rg_el ivg-i')
        for div in image_divs:
            meta = json.loads(div.text)
            if 'ou' in meta:
                self.put_task_into_queue(dict(img_url=meta['ou']))

Downloader

If you just want to change the filename of downloaded images, you can override the method
```
downloader.set_file_path(img_task)
```
The default names of downloaded images are counting numbers, from 000001 to 999999.

If you want to process meta data, for example save some annotations of the images, you can override the method
```
downloader.process_meta(img_task):
```
Note that your parser need to put meta data as well as image urls into task_queue.

If you want to do more with the downloader, you can also override the method
```
downloader.download(img_task, request_timeout, max_retry=3,
                    min_size=None, max_size=None, **kwargs)
```
You can retrive tasks from task_queue and then do what you want to do.

Crawler

You can either use the base class ImageCrawler or inherit from it. Two main apis are

crawler.__init__(self, img_dir='images', feeder_cls=Feeder, parser_cls=Parser,
                 downloader_cls=Downloader, log_level=logging.INFO)

and

crawler.crawl(self, feeder_thread_num=1, parser_thread_num=1,
              downloader_thread_num=1, feeder_kwargs={},
              parser_kwargs={}, downloader_kwargs={})

So you can use your crawler like this

crawler = Crawler(feeder_cls=SimpleSEFeeder, parser_cls=MyParser)
crawler.crawl(feeder_thr_num=1, parser_thr_num=1, downloader_thr_num=4,
              feeder_kwargs=dict(
                  url_template='https://www.some_search_engine.com/search?keyword={}&start={}',
                  keyword='cat',
                  offset=0,
                  max_num=1000,
                  page_step=50
                  ),
              downloader_kwargs=dict(
                  max_num=1000,
                  min_size=None,
                  max_size=None
                  )
              )

Or define a class to avoid using complex and ugly dictionaries as arguments.

class MyCrawler(Crawler):

    def __init__(self, img_dir='images', log_level=logging.INFO):
        ImageCrawler.__init__(self, img_dir, feeder_cls=SimpleSEFeeder,
                              parser_cls=MyParser, log_level=log_level)

    def crawl(self, keyword, offset=0, max_num=1000, feeder_thr_num=1, parser_thr_num=1,
              downloader_thr_num=1, min_size=None, max_size=None):
        feeder_kwargs = dict(
            url_template='https://www.some_search_engine.com/search?keyword={}&start={}',
            keyword=keyword,
            offset=offset,
            max_num=max_num,
            page_step=50
        )
        downloader_kwargs = dict(
            max_num=max_num,
            min_size=None,
            max_size=None
        )
        super(MyCrawler, self).crawl(
            feeder_thr_num, parser_thr_num, downloader_thr_num,
            feeder_kwargs=feeder_kwargs,
            downloader_kwargs=downloader_kwargs)

crawler = MyCrawler()
crawler.crawl(keyword='cat', offset=0, max_num=1000, feeder_thr_num=1,
              parser_thr_num=1, downloader_thr_num=4, max_size=(1000,800))

How to use proxies (optional)

A powerful ProxyPool class is provided to handle the proxies. You will need to override the Crawler.set_proxy_pool() method to use it.

If you just need a few (for example less than 30) proxies, you can override it like the following.

def set_proxy_pool(self):
    self.proxy_pool = ProxyPool()
    self.proxy_pool.default_scan(region='overseas', expected_num=10, out_file='proxies.json')

Then it will scan 10 valid overseas (out of mainland China) proxies and automatically use these proxies to request pages and images.

If you have special requirements on proxies, you can use ProxyScanner and write your own scan functions to satisfy your demands.

def set_proxy_pool(self):
    proxy_scanner = ProxyScanner()
    proxy_scanner.register_func(proxy_scanner.scan_file,
                                {'src_file': 'proxy_overseas.json'})
    proxy_scanner.register_func(your_own_scan_func,
                                {'arg1': '', 'arg2': ''})
    self.proxy_pool.scan(proxy_scanner, expected_num=10, out_file='proxies.json')

Every time when making a new request, a proxy will be selected from the pool. Each proxy has a weight from 0.0 to 1.0, if a proxy has a greater weight, it has more chance to be selected for a request. The weight is increased or decreased automatically according to the rate of successful connection.

API reference

To be continued.

Lightjohn/icrawler