Dataset Downloader

Preview

Dataset_downloader allow you to download large dataset from multiple list of url, from image-net for example. You can split the download into 2 folders, one for the training and one for the testing. File are save into their class name, perfect for model training. It looks something like that:

root:.
|
├───test
│   ├───accerola
│   ├───apple
│   └───lemon
├───train
│   ├───accerola
│   ├───apple
│   └───lemon

Installation

Simply install from pip:

pip install dataset_downloader

Config

Create a dataset.json file with the following content:

{
  "outputTrain": "...",
  "outputTest": "...",
  "ratio": ...,
  "classes": {
    "class1": [
      "http://url1",
      "http://url2"
    ],
    "class2": [
      "http://url1",
      "http://url2"
    ],
    "class3": "list_images.txt"
  }
}

• outputTrain: Output folder of the training images
• outputTest: Output folder of the testing images
• ratio: The ratio of training/testing images. 0.8 correspond of 80% of training images.
• classes: List of classes with their urls. Urls can be a list of url, a file containing a list of urls or an url containing a list of urls

An exemple of file on a windows computer:

  "outputTrain": "D:/dataset/train",
  "outputTest": "D:/dataset/test",
  "ratio": 0.8,
  "classes": {
    "accerola": [
      "http://tiachea.files.wordpress.com/2008/10/acerolas.jpg",
      "http://www.jardimdeflores.com.br/floresefolhas/JPEGS/A56acerola5.JPG",
      "http://farm2.staticflickr.com/1353/4602150961_177e096984_z.jpg",
    ],
    "apple": [
      "http://www.naturalhealth365.com/images/apple.jpg",
      "http://urbanext.illinois.edu/fruit/images/apple1.jpg",
      "https://www.aroma-zone.com/cms//sites/default/files/plante-acerola.jpg"
    ],
    "lemon": "list_images.txt",
    "watermelon": "https://gist.githubusercontent.com/johnrazeur/645787bc08a5aedd82da9573fbfa169a/raw/49cea1ee1438cecef8ac213b20f24e5ae02d4d78/watermelon.txt"
  }

Run

Simple call the dataset_downloader command:

cd yourdirectory
# You must create the dataset.json file before
dataset_downloader