Dataset creation

This is usually the most tedious part of a project, but do not worry, we have prepared the dataset for you!

  1. Follow the instructions in Setting up projects to create a new project. Name it "Tutorial"

  2. Click on "Add Images" and upload the following file: https://storage.googleapis.com/dp-public/platform/tutorial/cityscape/dataset.json. Use "Save As" if the JSON file opens up as a new tab of your browser. The explanations about the format can be found in Upload images via a text file but you do not need to worry about that for now.

  3. The upload of the dataset can take a couple of minutes but you can navigate in the platform meanwhile. Once done, you should see 3 "Views" like this:

This datasets is made of 3 views: "vehicules-lights-people", "context" and "lights".

The notion of "Views" is a particularity of our platform and allows you to create hierarchical datasets. You can learn more about views in Creating & organising views, but long story short: you can think of them as being sub-datasets providing annotations on the same set of images. Here we have:

  • The view "vehicles-lights-people": it contains bounding boxes of vehicles, traffic lights, pedestrians, bikers, etc...

  • The view "lights" is a child of the first view: it will show all the traffic lights annotated in "vehicles-lights-people". If you delete a traffic light bounding box from "vehicles-lights-people", it will be immediately propagated in the "lights" view (and the number of "images" it contains will be 1041 instead of 1042)

  • The view "context" is a "root" view, like "vehicles-lights-people", i.e. a view with no parent. We will use it to decide if the picture depicts an intersection or something else (that we call "road").

Try to explore those views. Try to change annotations, add some, remove some. Don't be afraid of clicking everywhere to change annotations on a few images (there is 1000 of them anyway). More documentation is available in Visualizing images and Annotating Images.

Once ready, let's move to the next section: