zalandoresearch/fashion-mnist

S3 bucket is providing different files from repo

Closed this issue · 1 comments

Noticed in this repo, but it happens directly in the browser for me as well.

MD5 hashes

manually downloaded from repo S3 links

7edbbf1fc824916c442268ac4dc845cd  - ./t10k-images-idx3-ubyte.gz.md5
b9859d5936603c782c6eb8dd14198360  - ./t10k-labels-idx1-ubyte.gz.md5
053aba987904a004d52cb333753041a3  - ./train-images-idx3-ubyte.gz.md5
7864864ad9592b0ffcc53c942eb67b24  - ./train-labels-idx1-ubyte.gz.md5

downloaded from repo directly

bef4ecab320f06d8554ea6380940ec79  - ./t10k-images-idx3-ubyte.gz.md5
bb300cfdad3c16e7a12a480ee83cd310  - ./t10k-labels-idx1-ubyte.gz.md5 
8d4fb7e6c68d591d4c3dfef9ec88bf0d  - ./train-images-idx3-ubyte.gz.md5
25c81989df183df01b3e8a0aad5dffbe  - ./train-labels-idx1-ubyte.gz.md5

hmm, but they should be consistent. You may try:

curl http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz | openssl md5

returns the same MD5 25c81989df183df01b3e8a0aad5dffbe as from the repo.