Tried playing with running numbers through that colab coe code but
was getting wrong answers. Downloaded the images from the project
and noticed that my numbers were small within the frame compared
to the original. Once I fixed that, all was good!!
So, we want jpg compression, on a 28x28x96 image...Around 600, 700 bytes.
* * *
Recaping the set-up:
One wants to get the code, which takes us to a GitHub page
One can pen the elements one at a time, and download raw: the arrow on the right.
We then create a Google Colab project to have access to a faster GPU and storage
facilities. The original uses 'cuda' but not everyone has an nvidia card on their
machine
To make it all manageable, I put in one cell the contents of one comment.
* * *
I'll be going back to Codemy pytorch later today, where we will be training
the model and evaluating it.
No comments:
Post a Comment