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