facefetch

A script to batch download sets of images matching specified criteria off Bing Images and organize each set in subfolders. Although this was originally used to gather training data for a face recognizer it can be utilized to accumulate any kind of image clusters.

Dependencies

facefetch is written in Python3 and requires lxml and Beautiful Soup 4 for html parsing

pip install lxml beautifulsoup4

Sample usage

git clone https://github.com/phoelix/facefetch.git
cd facefetch
mkdir {'Bud Spencer','Terrence Hill'}
./ffetch -n 10 --size large --type photo --face closeup --ftypes jpg --rename {dir}_{:02d}

This results in two folders each containing 10 jpgs of large face closeups, with file naming format below:

Options

As ffetch takes the names of all subdirectories as search query input, the only required option is the number of images to download (-n) into each folder.

Positional Arguments:

path                    absolute path to directory with image folders.
                        defaults to working directory

Optional Arguments:

  --help, -h            show help message
  -n                    number of images to download in each directory
  --size, -s            minimum image size. Values: all, small,
                        medium, large, xlarge. defaults to medium
  --min, -m  [w h]      minimum image size in pixel dimensions (width height)
  --type, -t            image type. Values: photo, clipart,
                        lineart, anim
  --face, -f            images with faces. Values: closeup, portrait
  --ftypes, -T [[...]]  limit to image file types. Any combination of: jpg,
                        png, gif, tiff, bmp, svg
  --rename, -F          specify format to rename image files.
                        Accepts python3 string formats, use {:d} for a running
                        number and {dir} for the subfolder name
  --verbose, -v         enable verbose output

Image sizes are pixel-defined as followed:

• small: max 200x200
• medium: 200x200 - 500x500
• large: min 500x500
• xlarge: min 1000x1000

Define a custom minimum size with --min (eg --min 300 400)

Notes

• Prefix foldernames with '.' to be ignored and not queried
• ffetch will skip folders containing >= the specified amount of images to download, allowing it to resume batch downloading after abort
• You will notice (in verbose mode) that some downloads fail due to HTTP 403/404/502... Errors. ffetch will collect an extra image for each failed download to make sure it always results in exactly n images for each subfolder.
• I have chosen Bing Images over Google Image Search as Google search result links are generated by javascript, which makes parsing slow and inefficient.