Hmm but the labels are automatically generated. Any model trained will be a second generation copy of the original labeling model.
Google releases the Open Images dataset, a strong alternative to ImageNet --larger-scale, with a better class set:https://research.googleblog.com/2016/09/introducing-open-images-dataset.html …
-
-
-
Even in the validation set only false positives were human verified, not false negatives.
End of conversation
New conversation -
-
-
Sadly there are zero images in the Open Images dataset, but there are 9 millions URLs. What a nightmare; let the bit rot begin.
-
To be fair: supporting research that can be replicated may not be the goal here. This would also explain the absense of a test set
- Show replies
New conversation -
-
-
I wrote download script on github.https://github.com/contaconta/Open-Images-downloader …
-
Beware of missing files and non-images. eg: https://c2.staticflickr.com/4/3046/2847433778_40cc71e590_o.jpg … https://c1.staticflickr.com/9/8545/8603563744_cb70fb5c68_o.jpg …
End of conversation
New conversation -
-
-
I hope FB never does that :) :)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.