Broken images
Thanks for the dataset, downloaded it in a first day after you've mentioned it on reddit.
Testing my tagging/captioning scripts found at least 3 images that are broken (partially downloaded).
As I'm not sure how the data was renamed in the new repo (opendiffusionai/pexels-photos-janpf) I'll leave this note here.
Broken files are:
/xbq/pexels-photo-1959103.jpeg
/xcu/pexels-photo-2036418.jpeg
/xdu/pexels-photo-933618.jpeg
Please check these files in the new repository.
I'll comment here if there will be more broken images although there's just over 7000 images left to re-tag so probably there will not be much.
I now see how files are named in opendiffusionai/pexels-photos-janpf:
The filenames are based on the md5 hash of each image
Unfortunately I've replaced these files already with freshly downloaded ones.
Looks like problem may be on my side, either unzipping broken or something similar:
Tagging found one more broken image, so md5sum it before re-downloading:
$ md5sum ./xfe/pexels-photo-2505670.jpeg
479b52afa7edb5b2a20b6d79234866a2 ./xfe/pexels-photo-2505670.jpeg
$
Download again from pexels and compare checksum to the list:
~/Downloads$ # Freshly downloaded image:
~/Downloads$ md5sum pexels-photo-2505670.jpeg
5b8f34b4d75e5385403c6a8e7bc00886 pexels-photo-2505670.jpeg
~/Downloads$ # Check against url-list-md5:
~/Downloads$ unzip -qq -c url-list-md5.zip | grep 2505670
5b8f34b4d75e5385403c6a8e7bc00886 https://images.pexels.com/photos/2505670/pexels-photo-2505670.jpeg
~/Downloads$
Thankyou for taking the time to give me feedback!
I encourage you to switch over to the new opendiffusionai/pexels-photos-janpf
There will be lots of additional features there, including
pre-calculated captions, and lists of subsets.
I'm going to close out this dataset here and just redirect people over
there now.