Computer vision one line commands

Less is more.

Setup

• Install docker on your machine.
• Clone this repo to your machine: git clone https://github.com/kungfuai/computer-vision-one-liners.
• Build it: ./bin/build.

Here we go

Train a binary image classifier, e.g. car in photo vs. not.

./bin/train <input_directory> [-o <output_directory>]

Your input_directory can look like this:

my_input_directory/
    car/
      1.jpg
      2.jpg
    no_car/
      1000.jpg
      1001.jpg

Train a multi-class image classifier, e.g. cat vs. dog vs. horse.

./bin/train <input_directory> [-o <output_directory>]

Your input_directory can look like this:

my_input_directory/
    cat/
      1.jpg
      2.jpg
    dog/
      1000.jpg
      1001.jpg
    horse/
      2000.jpg
      2001.jpg

Train a multi-label binary image classifier, i.e. photo2tags.

./bin/train <input_directory> [-o <output_directory>]

Your input_directory can look like this:

my_input_directory/
    labels.txt
    images/
      1.jpg
      2.jpg
      Photo 3.jpg
      a_sub_folder/
        4.jpg
      ...

The labels.txt file has the labels/tags per image. Each line starts with the relative path of an image, then its tags separated by a space:

images/1.jpg cat blue_sky my-favorate
"images/Photo 3.jpg" cat
imags/a_sub_folder/4.jpg dog indoor