Let's do it with an example !

Objectives

We want to build a Dataset in order to train a model to differentiate people from vehicles in an aerial view, but unfortunately we do not have a custom Dataset for this precise task.

Hopefully we do have a Dataset that could be a great fit with some modifications.

We have a Dataset called VizDrone Dataset (available on our Dataset Hub), that contains 6470 pictures, and more thant 30k + objects annotated with the given repartition.

Creation the right version of this Dataset

Ok, so let's create a new version of this Dataset in order to perform operations on it.

First let's go the VizDrone Dataset overview and select all assets

Then click on new version :)

You can now create your new version :) here we called it tutorialVersion

The version creation can be quite long, but don't worry you can change page of grab a cup of tea !

Configure your new labels

Ok, now we can declare our 2 new classes people and vehicule in order to merge all the vehicule related annotations and human annotations in this 2 classes.

You should now see

No annotations yet ... well not for long, let's go to import annotations

Merge your annotations

Ok, so the import page is quite simple, juste select the new class where you want to import annotations

Here we want to import and the pedestrian and people annotations into the new people class

Then click on merge :)

You can do the same thing for vehicule :)

Once your done you will see the summary of the ations to perform here

Just click on execute instructions and your done :)

Now we have a balanced dataset with the classes that interest us :)