Open Images

https://storage.googleapis.com/openimages/web/index.html

https://github.com/cvdfoundation/open-images-dataset

root=data/open_images
mkdir -p ${root}/annotations

for i in {0..9}; do
    echo ${i}
    aws s3 --no-sign-request cp s3://open-images-dataset/tar/train_${i}.tar.gz ${root}
done
for i in {a..f}; do
    echo ${i}
    aws s3 --no-sign-request cp s3://open-images-dataset/tar/train_${i}.tar.gz ${root}
done
aws s3 --no-sign-request cp s3://open-images-dataset/tar/validation.tar.gz ${root}
aws s3 --no-sign-request cp s3://open-images-dataset/tar/test.tar.gz ${root}

for f in ${root}/*.tar.gz; do
    echo ${f}
    tar -zxf ${f} -C ${root}
done

data/open_images/
├── train_0
|   ├── 000002b66c9c498e.jpg
|   └── ...
├── ...
├── validation
|   ├── 0001eeaf4aed83f9.jpg
|   └── ...
└── test
    ├── 000026e7ee790996.jpg
    └── ...

The training split contains 1,743,042 images, the validation split contains 41,620 images, and the test split contains 125,436 images.

Open Images v6

wget https://storage.googleapis.com/openimages/v6/oidv6-train-annotations-bbox.csv -P ${root}/annotations