Hasty offers a wide variety
of manual and semi-automated tools to get you started with your
annotation efforts. Depending on your project, different tools might be
preferable. More on this here.
Let's start by reviewing our manual tools.
These are tools are all available when you start a new project. They offer no assistance or AI functionality.
This tool is used to create bounding boxes in two clicks. It is only used when working on object detection annotation tasks.
The polygon tool is used to create polygons with 2 or more clicks. It
is mainly used for instance segmentation but can be used for semantic
segmentation and panoptic segmentation sometimes.
The brush allows you to draw shapes as you would when using a
graphical editing software (Microsoft Paint or Photoshop, for example).
It can be used for any type of segmentation annotation task.
Image tags are what we call a single classification added to an
image. It is only used when working on image classification tasks.
Label attributes are additional pieces of information concerning an
annotation beyond the class. For example, you might have labeled an
apple, but you also want to know which type of apple it is, and what its
"freshness" is. For this, we use attributes. Think about attributes as a
way to add additional metadata to your annotations.
Semi-automated tools are also available from the start of any
project. These tools use pre-trained AI models to speed up annotation
from the first image.
DEXTR is an AI-powered tool where you put in the north, east, south,
and west extreme points of an object you want to annotate. The AI model
will then look for a shape between those points. It is mainly used for
instance segmentation, but can also be used for semantic and panoptic
ATOM is an AI-powered segmentation tool that allows you to annotate
in one click. Simply click in the middle of the area/object you want to
annotate. Get a suggestion back. Then, you can manipulate the suggestion
by adding more clicks. In our testing, we found that we could do
complex annotations in 2-5 clicks on average.
Box to Instance is a bit of an outlier as it is a tool for converting
bounding boxes to instances. It does so by taking any bounding boxes
present on an image and then look for shapes inside of those boxes. Its
main utility is to assist you if you want to migrate a project from
object detection to instance segmentation or panoptic segmentation.
In some cases, you might want to convert polygons to masks or vice versa. We have built functionality for those cases.
In other cases, you might find yourself with many annotations that
should be one, uniform, one. For this, we have merge functionality.