pixplz

Fetch sample images for a search term, without fuss.

Install

pip install pixplz

For faster results, but with more dependencies, you can do:

pip install pixplz[parallel]

If you have a version of ssl that is giving warnings, and they annoy you, do:

pip install requests[security]

Options

$ pixplz koala
5 images downloaded (1 duds skipped)
$ ls
koala_000000.jpg  koala_000002.jpg  koala_000004.jpg
koala_000001.jpg  koala_000003.jpg
$ pixplz -h
usage: pixplz [-h] [--prefix PREFIX] [--format FORMAT] [--parallel PARALLEL]
              [--count COUNT]
              term [term ...]

positional arguments:
  term                 search term

optional arguments:
  -h, --help           show this help message and exit
  --prefix PREFIX      prefix for downloaded images, _NNNNNN.jpg will be added
  --format FORMAT      format for naming images, for example 'foo/bar%06d.png'
  --parallel PARALLEL  number of images to load in parallel (requires
                       pixplz[parallel])
  --count COUNT        target number of images to load (fewer may be loaded)

Source of images

Currently, this grabs a few images from ddg to use as casual training data. After poking around for terms of service, I still don't know how ddg feels about this. Will switch to other sources of public image data as needed.

Clearly, images from a search engine are not going to be solid ground truth. I use pixplz when near enough is good enough. For example, recently I used an amalgam of text, slogan, logo against wall, sky, night when training a network for billboard segmentation.

Now with audio!

mp3plz cooking

You now have some mp3s of cooking sounds, from freesound.org, for use as casual training data. I don't know how freesound feels about this. Will switch to other sources of public audio data as needed.