a) Download the the images of Visual Genome Part1 and Part2. Unzip and place all images in a folder data/vg/images/
b) Download the annotations of VG (in COCO-format) and unzip it in the data/ forder.
a) Download the original annotation of Open Images V6 (oidv6/v4-train/test/validation-annotations-vrd.csv) from the offical website.
b) Download the OIv6 images provided by Rongjie Li and unzip it.
c) Change the paths in process.py. The images will be renamed (since the COCO tool only supports numeric image names) and the COCO-like annotations will be produced. You can also download the annotations with the link, but still need to rename the images.
d) Put the renamed images in the data/oi/images/ and the annotations in data/oi/.