Thanks for sharing
Hi @jkorte,
I’d say probably the easiest way to understand how this works is to take a look at this notebook which demonstrates how this technique can be used.
The timm package is truly great and also quite often updated with new architectures or adjustments to the existing ones.
make sure to install from master to make use of the cutting-edge SOTA models as soon as it’s added:
pip install git+https://github.com/rwightman/pytorch-image-models.git
for image segmentation, dose fastai library have a labeling tool for it?
How do I get the same presentation setup Jeremy is using in jupyter? I have not seen that functionality to switch between powerpoint and code
I had the same question, it’s pretty neat
No I haven’t seen a labeling tool for this. There are setups for image classification however
FastAI does not have a labelling tool, there a bunch of opensource ones available.
Why is a DataBlock
called a DataBlock
? I guess it’s a pretty hard thing to name, but I feel like a DataPreparation
would be a more intuitive name.
Can we get access to Jeremy’s slides before the course video is released? Or is the same content covered in the lesson videos/book/etc?
Notebooks are great:
Book written with notebooks - GitHub - fastai/fastbook: The fastai book, published as Jupyter Notebooks
Library written with notebooks - fastai/nbs at master · fastai/fastai · GitHub (using nbdev)
Blogs written with notebooks - https://fastpages.fast.ai
Presentations in notebooks- RISE — RISE 5.7.1
There are some variations in this iteration of the course versus the previous version. But the core concepts are the same.
I will ask Jeremy to post if he feels appropriate.
Feels like Google/Deep Mind tries to rethink their ML stack and use JAX for some recent publications, like Gopher?
Datablock name? IIRC fastai called it a Databunch before settling on Datablock. Pytorch natively has a “Dataset” to feed dataloaders. A datablock is essentially a Dataset with a bunch of helpers.
EDIT: Jeremy corrects me, "DataBunch
was renamed to DataLoaders
. DataBlock
isn’t much related to a Dataset
. It’s a builder for DataLoaders
."
I have misremembered from 2yr ago. Good thing I am doing the course