Datasets:
Final Dataset Size
Hi! Thank you for the great work, I would very much appreciate an information of what is the size of the final datasets and info on what is in the single zip files:) - are they grouped by classes? Can one pick a subset of the train_{i}.zip files?
Hi :)
The zip files contain diffs to ImageNet1k, as I personally don't have the rights to the ImageNet dataset. These are the masks of foreground objects, as well as the pixel-wise differences of how to fill in the foregroud objects to create neutral backgrounds. Thus, you additionally need the ImageNet dataset. For information on how to handle these objects, see the Local Installation section and apply_patch.py in this repo: https://github.com/tobna/ForAug?tab=readme-ov-file#local-installation
Indeed, the train_{i}.zip files are grouped by classes, so you can pick a subset of files and get a subset of ImageNet.
The final dataset size (without the zips) will be ~75G for foreground and background images, or ~55G if you create the optional zip files (as described in the other repo)