Feature/dg 979 support classification

[BloodAxe]

• Adds support for classification task on public API level
• Adds simple features (class distribution)
• Improve ask_user method to use ipython widgets instead of stdin when launched from notebook to allow users to provide input

[shaydeci]

• Adds support for classification task on public API level
• Adds simple features (class distribution)
• Improve ask_user method to use ipython widgets instead of stdin when launched from notebook to allow users to provide input

In SG we kept colab links rather then having the ipynb files inside the repo, I think its better that way (in case changes are introduced we will always need to update in 2 places...)

[BloodAxe]

• Adds support for classification task on public API level
• Adds simple features (class distribution)
• Improve ask_user method to use ipython widgets instead of stdin when launched from notebook to allow users to provide input

In SG we kept colab links rather then having the ipynb files inside the repo, I think its better that way (in case changes are introduced we will always need to update in 2 places...)

Well I can argue that this break "atomic change" concept, that you have to make a DF release first, then go over all colab files and update them by hand. From my perspective having them under git is easier to maintain. These are meant to be "quick usage examples" that people can quickly refer to.
Maybe for deep dive colabs as we did for YOLO-NAS it make sense to store them elsewhere...
But honestly - location of notebook files is the smallest question I have about this PR ):

[Louis-Dupont]

Looks good, I left some comments inline but nothing too important.

As a bonus, if you can add a minimalistic end-to-end test with a dummy dataset (which can be built on the fly with random images and only a couple of random classes.
We don't have such tests for segmentation and detection but I think we should asap.

[ofrimasad]

a few minor comments

[Louis-Dupont]

Looks like the notebook still includes this plot

[Louis-Dupont]

LGTM

[Louis-Dupont]

Note: We agreed on adressing the changes concerning the denormalization in a following RP