Right, but are there other values that could be? Like, is there a sensible way you could have a 1/3 stride? I’m not sure what it’s a half of, if that makes sense.
Is the intuition here that the input image is not the best location to perform our convs?
@rachel Hi, where can we find the notebook that Jeremy is showing? the imagenet processing in parallel notebook
2 Likes
I’d like to know this too.
I think this is the DeViSE paper. Not spam!
1 Like
I think it hasn’t been added to platform.ai yet, Jeremy will add it after class
2 Likes
I tried asking for the ImageNet dataset about a month ago. No response.
Also, they don’t let you use public email services either (gmail, hotmail, etc.) in the request form (so they say).
1 Like
How will you encode the word vector if it is compound word like “jumbo jet”?
got it, thanks!
This might help!
1 Like
As a different way of encoding labels is there some way to label heirarchical structure?
@mgarrus wow… long answer. thanks. I’ll check it out.
Can we use Doc2Vec to transform phrases to vectors?
Anything can be a vector these days …
2 Likes
5 Likes
How could you get the relation between a pug and a dog and the photo of a pug and its relation to the bigger dog category?
1 Like
When resizing what if we lose some pixels?
Why do we want to randomize the filenames? doesn’t keras ImageDataGenerator. flow_from_directory shuffle the data for us if we want it to?
1 Like
Where is imagenet_process-full-Copy1.ipynb?
4 Likes
Yes,
Please @jeremy, where could we get the “Imagenet Processing in parallel” notebooks ?
It is not in the https://github.com/jph00/part2 link.