Skip to content

Latest commit

 

History

History
72 lines (64 loc) · 2.07 KB

DATASETS.md

File metadata and controls

72 lines (64 loc) · 2.07 KB
Raw

How to install datasets

The datasets we use are listed below:

  • Visual-genome Link
  • HAKE Link
  • Visual Attributes in the Wild (VAW) Link
  • SWiG Link

As the images in the VAW dataset come from the Visual Genome dataset, we only need three picture folders. The file structure looks like below:

$DATASET/
|–– hake/
|–– swig/
|–– vg/

HAKE

  • Create a folder named hake/ under $DATASET/.
  • Create hake_images_20190730/,hake_images_20200614/,hcvrd/,hico_20160224_det/,openimages/,pic/ and vcoco/ under hake/.
  • Download the images from the dataset link.
  • Put the images in the corresponding folder.

The folder tree looks like:

hake/
├── hake_images_20190730
│   └── xxx.jpg
├── hake_images_20200614
│   └── xxx.jpg
├── hcvrd
│   └── xxx.jpg
├── hico_20160224_det
│   └── images
│       ├── test2015
│       │   └── xxx.jpg
│       └── train2015
│           └── xxx.jpg
├── openimages
│   └── xxx.jpg
├── pic
│   └── xxx.jpg
└── vcoco
    ├── train2014
    |    └── xxx.jpg
    └── val2014
        └── xxx.jpg

SWiG

  • Create a folder named swig/ under $DATASET/.
  • Download the images from the dataset link.

The folder tree looks like:

swig/
└── xxx.jpg

VG & VAW

  • Create a folder named vg/ under $DATASET/.
  • Create VG_100K/ and VG_100K_2/ under vg/.
  • Download the images from the part1 and part2.
  • Put the images in the corresponding folder.

The folder tree looks like:

vg/
|–– VG_100K/
    └── xxx.jpg
|–– VG_100K_2/
    └── xxx.jpg