Open PGijsbers opened 3 weeks ago
From a related item: We are currently with some prototypes of downloading bucket content with many images, however there are many things left unspecified (and unsupported by packages), e.g.:
dataset upload (adding auxiliary files in general) how should the files be zipped/unzipped, and how can we know at download time how to resolve the paths? how should we store metadata/which metadata should be store (different types of tasks, bounding boxes, segmentation masks, etc.) add the relevant documentation More generally should extend to parquet file describing other files (images, audio, video).
In general, image datasets currently consist of a header table with a directory of files. So a "File Dataset" may be more apt.