Add ClassificationDataset API

[capjamesg]

Description

This PR introduces a new ClassificationDataset API for use in working with classification datasets using supervision. Users can:

• Create a ClassificationDataset object;
• Save data in the multiclass classification folder, and;
• Load data from a dataset structured as a multiclass classification folder.

Type of change

• New feature (non-breaking change which adds functionality)

How has this change been tested, please provide a testcase or example of how you tested the change?

I will add a test case once we have discussed the API.

Docs

I will add the requisite docs once we have discussed the API.

[SkalskiP]

A summary:

• Classifications class should not be in the datasets package.
• Splitting dataset is not a responsibility of as_multiclass_folder_structure.
• Please do not use+ "/" + to build file paths.
• Classification API should be consistent with detections API. When users want to load train, test, and validation subsets, they should create three separate instances of ClassificationDataset. You can find more info in one of the code comments.
• Running as_multiclass_folder_structure and from_multiclass_folder_structure should result in precisely the same dataset.
• Make sure to run isort --profile black supervision/ and black supervision before commit.

[capjamesg]

@SkalskiP I have responded to a lot of your feedback in my latest commit. Please review. I am still confused about this comment:

Splitting dataset is not a responsibility of as_multiclass_folder_structure.

Can you outline in pseudocode the expected way this function would work?

[SkalskiP]

@capjamesg test, train, and valid sets will be represented by separate ClassificationDataset objects. And that's why there is no reason for you to split the dataset into parts inside as_multiclass_folder_structure. It will be already split by logic operating on a higher level. as_multiclass_folder_structure is here to save the dataset on the hard drive and do nothing more than that. Here is an example of the usage of the Classification API

ds = ClassificationDataset(...)
ds_train, ds_test = ds.split(split_ratio=0.8)
ds_test, ds_val = ds_test.split(split_ratio=0.5)

ds_train.as_multiclass_folder_structure(...)
ds_test.as_multiclass_folder_structure(...)
ds_val.as_multiclass_folder_structure(...)

Notice that split call is not inside of as_multiclass_folder_structure, but outside.

[capjamesg]

My latest commit incorporates all your feedback. I left a comment in response to one of your comments re: the class_id [0] issue.

[capjamesg]

My latest commits add:

1. Documentation for the new methods;
2. Unit tests for get_top_k;
3. Linting for all changes, and;
4. A solution for the if self.annotations[image].class_id[0] == self.classes.index( issue.

[capjamesg]

I have incorporated all of your feedback into my latest commit. We now have five test cases for the get_top_k() function.