If you have used graphical editing tools or other annotation tools, our annotating environment should be familiar. For newcomers or users that wonder where they can find a specific function, we put together this quick tour of the interface that explains in more general terms what different parts of the interface does and why they are there.
At the top of the interface, you can find the navigation bar. Here, you can navigate between images, access the project menu, and change the image’s status, as well as accessing various menus. To go to another image, click the arrows. To change the image status, click on the current image status to access a drop-down menu, in which you can specify a new status.
On the left side, you have the toolbar. This is where you can access all of the tools that you can use to annotate.
You use a tool by first clicking on it's icon. When it's selected (which you'll know as the tool icon will have a darker square around it) you can use the tool on your image.
On the right side, you have the label class list where you can select which label class to assign to objects as well as see which objects already have a specific label class.
The label class list is split up in two parts, foreground and semantic classes.
What's really important to understand here is the concept of "active class". As Hasty is a tool built for speed, we decided that we wanted to allow users to create many different annotations with the same class as quickly as possible. To that, when you select a specific class as an "active class" by clicking on the hedgehog icon to the left of class name, every new annotation being made will have that class assigned to it.
However, "active class" just applies to new annotations. To edit the class of an existing annotation, first select it using the move/edit tool. The current class of the annotation will be highlighted in blue.
To change the class, click on the desired class in the class list - please note that you don't click on the hedgehog symbol as this will change the active class, and not edit the class of the selected annotation.
You will also find the history list directly underneath the label class list. In the history list, you can see previous actions that you have performed when working on a specific image. You can also undo actions that you are not happy with here.
In the history list, new events are added to the bottom of the list. To change back to a previous state, just click on the event from which you want to roll back to. Please note that wherever you click in the list, all events that took place after this specific event will also be undone.
There are two parts of the interface that are not directly visible when you are opening a new project. First, we have the tool settings bar, that only will be shown after you have selected a tool. Here, you can modify how the tool works in various ways. For example, if you select the magic wand tool, you will see this menu at the top of the screen:
Please note that any changes made in the tool settings bar require you to perform an action afterwards for the changes to take effect. For example, if you select an area using the magic wand, and then decide to increase the tolerance, this change will not be reflected until you make a new selection using the magic wand.
Secondly, we have the image gallery that can be accessed by clicking the arrow icon at the bottom of the screen. The image gallery is an easy to access view of all images in the project.
It can be accessed by pressing the two arrows pointing up at the bottom of the screen, and can be hidden by pressing the two arrows pointing down on top of it when it's open.
You can also filter the images being shown here by dataset and image status. Just go to the filter you want to adjust at the top right of the image gallery, click on it, and pick what filter should apply.
Furthermore, you can search on an image name by filling in the search field that can be found at the top right of the image gallery.
That leaves us with the centre of the interface where you can see your image and perform actions on that image. We call this area the work area.
In the current version of Hasty, we have 8 tools that you can find in the toolbar. Those are:
With this tool, you can move or edit already existing objects. You also need to use it to relabel objects after they have been created.
The polygon tool allows you to create more complex, polygon-based shaped objects.
This tool allows you to create bounding boxes, meaning simple rectangle-shaped objects. It also has a built-in smart assistant, pattern matching.
The contour finds potential objects in your image by looking for outlines of objects where the contrast is high. You can then easily convert these contours to objects by clicking on them.
The magic wand lets you select a selection area by finding similarly coloured pixels to the one you choose by clicking in the image.
The Brush and the eraser tools are used to create (using the brush) and remove (using the eraser) masks or edit masks.
Interactively extract objects in the foreground from the background with the help of strokes. Useful for selecting and creating complex objects.
DEXTR is an AI-powered tool that will be activated a couple of minutes after you create a new project. DEXTR can be seen as a collaboration tool between you and AI. You start by marking four (or more) edges of an object. The DEXTR-model will then "look" at those four edges and try to find an object it recognizes inside of between them.
This AI-powered tool shows you potential bounding boxes in your image. It will be activated couple of minutes after you have labelled 10 images with bounding boxes and set the image status of those 10 images to "done" or "to review".
These AI-powered tools shows you potential instances in your image. They will be activated a couple of minutes after you have labelled 10 images with polygons or masks and set the image status of those 10 images to "done" or "to review".
If you want to use the semantic segmentation assistant, know that you will need to create at least one semantic class and use it on at least one annotation, otherwise the tool will not train.