Modules

Datasets

Datasets.downloaderMethod
This function takes a url and a destination and downloads the file there with a name specified.

url: The url to be downloaded
dest: Path
fname: File name
```julia
downloader("http://www.julialang.org/", "/tmp/","index.html")
```
source
Datasets.get_dataMethod

Main function to download a standard dataset. Takes input as name from the above list, the destination and a boolean to check if the user wants to extract the data. The filename is inferred from the destination and the url.

get_data("mnist", "/tmp/", 1)

If not found, the list of available datasets is returned.

source
Datasets.getextMethod

Helper function to get the extension of the datasets. Takes into account patterns like .tar, .tar.gz , .zip etc..

source

Classification

classification.encoderMethod
  • Return a label encoded vector from 1..number of unique vals
  • Also returns a copy of the labels for later use, use as encoded, _ if you do not care about re encoding it again
source
classification.examplelabellerMethod
  • A function to show a custom labeller function as an example
  • This splits the path based on "/" and returns the lower case of the last element
source
classification.from_csvFunction
  • Read paths from a csv file along with labels given path column name and label column name
  • Further can be passed into labellers and encoders
  • Just a convinence function really
source
classification.from_folderMethod
  • Identify the number of classes in the data. Return an array with all image paths and class labels
  • Does not actually load the images. Just returns paths
  • Assuming heirarchy of classes to be:
    • Main Folder
      • class1 -img1 -img2...
      • class2 -img1 -img2...
source
classification.labelFromPatternMethod
  • Takes paths generated from load_classes as input
  • Takes a function defining a custom label
  • eg: labeller(x) = split(x, "/")[-1]
  • Returns labels based on function specified
source
classification.loadimFunction
  • Input path -> Read an image -> Convert to array -> Resize to given dimensions -> Convert to float
source

Available Datasets

DatasetLink
mnistSource
cifar10Source
cifar100Source
birdsSource
caltech101Source
petsSource
flowersSource
foodSource
carsSource
imagenetteSource
imagenette320Source
imagenette160Source
imagewoofSource
imagewoof320Source
imagewoof160Source
imdbSource
wikitext103Source
wikitext2Source
wmtSource
agSource
amazonSource
p-amazonSource
dbpediaSource
sogouSource
yahooSource
yelpSource
p-yelpSource
camvidSource
pascalSource
hmbdSource
ucfSource
kinetics700Source
kinetics600Source
kinetics400Source