Comment Comment by dnth The authors at fastdup ran an analysis on LAION 400M and Imagenet21K. Here's what they found.LAION 400M> 60M duplicates. > 962K broken images. > Various label discrepancies.ImageNet21K> 1.2M duplicate images. > 104K train/val leak.fastdup GitHub repo - https://github.com/visual-layer/fastdup parent dnth 3 years ago
LAION 400M
> 60M duplicates. > 962K broken images. > Various label discrepancies.
ImageNet21K
> 1.2M duplicate images. > 104K train/val leak.
fastdup GitHub repo - https://github.com/visual-layer/fastdup