Add it to a GitHub README or profile β the card updates as the ranking changes:
Easily turn large sets of image urls to an image dataset. Can download, resize and package 100M urls in 20h on one machine.
Easily compute clip embeddings and build a clip retrieval system with them