The original dataset provided by Google, here, consists of 'Image URL - Caption' pairs in both the provided training and validation sets. I have to work on an image captioning project and wanted to download all the images from the listed URLs. But, before doing that I just wanted to know the total size of the ~3,000,000 images listed in the dataset.
PS: On that note, I was wondering whether I have to download the entire dataset for training or is there a way I can train my network in an online setting where a thread concurrently downloads the images required in the current minibatch. Is there a way to do that in Tensorflow/Keras or some other framework?