A tool to download and format MS COCO dataset for multilabel image classification
It outputs a .h5 file that contains the following:
- data_types: 'train' and 'val'
- cats: names of the 80 categories (replace x with any data type)
- x_images: flattened images (not preprocessed, except for the few grayscale images that have been converted to RGB)
- x_shapes: shapes of the images, to reshape the flattened images
- x_names: file names of the images
- x_label: a one-hot integer vector of labels