Skip to content

Latest commit

 

History

History
81 lines (55 loc) · 2.37 KB

README.md

File metadata and controls

81 lines (55 loc) · 2.37 KB
Raw

FaceDatasets

Build Status

Package for making access to popular face datasets easier.

General usage

load_images(<dataset_name>, <dataset_params>...)  # faces
load_shapes(<dataset_name>, <dataset_params>...)  # face landmarks
load_labels(<dataset_name>, <dataset_params>...)  # provided labels

For example, to load CootesDataset, enter:

load_images(CootesDataset)

Note, that all methods return iterable objects: for small datasets they are just arrays, for larger iterators are returned instead. You can always materialize them using:

collect(load_images(...))

Available datasets

Cootes images

CootesDataset contains images from Tim Cootes' work on active appearance models. These images come prepacked, so you can use them for testing.

Supported functions:

  • load_images(CootesDataset)
  • load_shapes(CootesDataset)

Cohn-Kanade+ dataset

CKDataset contains images from Cohn-Kanade+ Expression Database. To install this dataset, download it from this page and unpack into a directory of your choice. Example of expected directory layout:

$ tree -L 2
.
├── cohn-kanade-images
│   ├── S005
│   ├── S010
│   ├── S011
│   ...
│   └── S999
├── Emotion
│   ├── S005
│   ├── S010
│   ├── S011
│   ...
│   └── S999
└── Landmarks
    ├── S005
    ├── S010
    ├── S011
    ...
    └── S999

Supported functions:

  • load_images(CKDataset, datadir, opts...)
  • load_shapes(CKDataset, datadir, opts...)
  • load_labels(CKDataset, datadir, opts...)

where datadir is base dir for CK dataset and labels are numbers representing 6 basic emotions + neutral facial expression.

Options:

  • start - image index to start with
  • count - number of images to return
  • indexes - concrete indexes to return (start and count are ignored)
  • resizeratio - resize image by this value

Cohn-Kanade+ (max only) dataset

CKMaxDataset - same as Cohn-Kanade+ dataset, but contains only images with maximally expressed emotion (~500 images). Only resizeration option is supported, though.

TODO: KaggleFERDataset TODO: PutFrontalDataset