Captcha dataset

This object implements a dataset using the torch::dataset() framework. It loads all the images in torch tensors, as well as the labels.

Usage

captcha_dataset(
  root,
  transform_image = captcha::captcha_transform_image,
  transform_label = captcha::captcha_transform_label,
  augmentation = NULL
)

Arguments

root

(string): root directory where the files are stored

transform_image

(callable, optional): A function/transform that takes in an file path and returns an torch tensor prepared to feed the model. By default, uses the captcha_transform_image() function.

transform_label

(callable, optional): A function/transform that takes in the file paths and transform them. By default, uses the captcha_transform_label() function.

augmentation

(function, optional) If not NULL, applies a function to augment data with randomized preprocessing layers.

This is an object of class dataset_generator created using torch::dataset() function. It has a initialize() method that takes a directory containing the input images, then assigns all the information in-memory with the array data structure for the response variable. It also has a .getitem() method that correctly extracts one observation of the dataset in this data structure, and a .length() method that correctly calculates the number of Captchas of the dataset.

The function calculates the vocabulary based on the identified values in the dataset.

Examples

if (torch::torch_is_installed()) {
  annotated_folder <- system.file(
    "examples/annotated_captcha",
    package = "captcha"
  )
  suppressMessages({
    ds <- captcha_dataset(annotated_folder)
  })
  # gets the first item (the only item in the example)
  # returns a list with x and y torch tensors.
  ds$.getitem(1)
}