Interesting result - VGG 2x faster on @pytorch vs keras + @tensorflow https://medium.com/@vishnuvig/transfer-learning-using-pytorch-part-2-9c5b18e15551 …
Seems to mainly be due to preprocessing time
-
yeah... can't really blame'em. all of my code use feed_dicts too. :-)
The overhead of feed_dict is small. However, making input preprocessing blocking instead of having it happen in parallel to execution ->
-
-
yeah, for the nlp stuff i was focused on before i couldn't care, was just a bunch of indexes. & then i started playing with images....
-
I also haven't touched pixel images for a while. Working with vector images feels closer to NLP, where bottleneck is probably elsewhere.
- Show replies
New conversation -
-
-
I am using keras.preprocessing.image ,is there a way I can use parallel execution in it. Or should I use directly tensorflow directly in it
-
Just use `pickle_safe=True` and `num_workers=4` (for instance) in your call to `fit_generator`. That is all.
- Show replies
New conversation -
-
-
Doesn't his comparison use the generator provided by keras?
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.