Skip to content
Enlarge training dataset by searching images with specified keywords in google and download the presented images
Branch: master
Clone or download
Latest commit e823a47 Jan 5, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
imgs add text to gif Mar 23, 2018
.gitignore Small changes Sep 30, 2017
LICENSE Modify license Sep 28, 2017 update readme Apr 16, 2018 change order of Exception Oct 2, 2017 fix issue #2 Jan 5, 2019 fix issue #2 Jan 5, 2019


This repository can download images from google with specified keywords


  • python 3.5
  • selenium 3.6.0
  • Firefox
  • geckodriver

Firefox and geckodriver are required by selenium, and Firefox 55 or greater is recommended, cause geckodriver support that best. As for geckodriver, just download the latest version of geckodriver from here, then add it to the PATH environment variable.

Here is an example of downloading with selenium using 2 processes

Details and Configuration

Two Methods

Two different methods are provided

  • download_with_urllib
  • download_with_selenium

download_with_urllib is to download with just package urllib, but due to the limit by google, each searching query can download at most 100 images

download_with_selenium is to download with package selenium and urllib, with selenium, we can directly search and scroll in the browser, so we can get more than 100 images for each searching query

Both of the above methods support downloading with single process or mulitple processes, and download_with_selenium will firstly store the actual links of images in a file, then download the actual images with the file, while download_with_urllib will directly download all the images since the number if small

Specify main_keywords and supplemented_keywords in the code, each main_keyword will join with each supplemented_keyword to become a searching query, and one directory will be created for each main_keyword to store the related images.The following image is a simple example

Download With Time Limit

As to the script, it is a replacement of the method download_images in script, because the method download_images will always block due to network issue, so I add restriction that each http request can cost at most 10 sceonds, and that is what does.

Pay attention that the time-limited strategy is to use the signal that system provides, and here the SIGALRM in unix-like system is adopted, so this script should run with unix-like system rather than Windows. However, the network blocking thing happened when I ran the script in my network, but not sure whether this will happen in yours, so you can test with firstly, if the downloading task blocks, change to the script

More details about the repository can be obtained in this blog

You can’t perform that action at this time.