Creating a Basic Shape Dataset
In this end-to-end example, we will create a basic shape dataset and import annotations already in COCO format. Once imported, we will adjust some annotations and export the data back to the COCO format.
You will need to download the Shapes Dataset.
- Installing COCO Annotator
- Creating a Dataset and Categories
- Importing Images
- Importing COCO Annotations
- Exporting Dataset
Please follow the instructions on getting started before continuing.
Once you have successfully launch COCO Annotator and navigated to the home page you should be greeted with the following:
Before proceeding, check the top right corner to make sure you see Done. This indicates that the client has successfully connected to the backend/database.
Before we are able to annotate we must create a dataset with some categories. A Dataset contains images and is a way of separating different types of training data. If you would like more information about datasets or categories, you can click the blue information icon located beside the title.
To create a dataset start by clicking the green Create button located under Dataset tile. A popup with ask for a name as well as categories (no categories will be listed, we will come back later and add some to our dataset). Give it a unique name, and take note of the directory (found under Folder) and click Create Dataset.
Navigate to the Categories tab found on the navigation bar, once again you will be presented with a create button. Here you can create as many categories as you would like.
Go back to the Datasets tab and click the Edit options found in the dropdown of the dataset card. Your category should be displayed in the categories window. From here you can create and select as many categories as you want.
Once you click done you have added a category to the dataset and it should be displayed in the cards footer.
In this section we will be looking at using the file watcher methods to import images into a dataset. For windows users the process is slight different and a breif overver can be found here. While the annotator is running the datasets directory is being monitored for file events. Whenever an image is deleted, created or moved the corresponding event is reflected in the database.
Advantages:
- Easy to import large datasets
- No file size limit
- Simple
Disadvantages:
- Requires access to the machine running annotator
- Complicated for windows users
Navgivate to the datasets directroy located at the root of this project. Inside this folder you should see folders with the names of the datasets you have created. From here simply place images inside the corrsponding folder to add them to the dataset. Subfolders are highly recommended as they can be used as a filtering methods in the client UI.
The annotator will start adding the images to the dataset, give it a few seconds before refeshing the client.
WIP.