Do instance segmentation 10x faster

Hasty is the world’s first annotation tool using AI to make you faster when annotating. Our human-in-the-loop system allows you to go from spending weeks doing annotations to days, or even hours.

An image of cars that have been instance segmented using Hasty.ai
An illustration of a brain
Powered by Artificial Intelligence
An illustration of an arrow pointing upward
10X faster annotations
An illustration of a shrinking circle
Reducing human error to 0
An illustration of an open padlock
Try for free

How we make you 10x quicker

Our instance segmentation toolbox

We have numerous tools for creating instances, both manual and AI-powered.

Start working faster today

Learn more about instance segmentation

What is instance segmentation?

Instance segmentation is an annotation method that is used when you want to identify every instance of an object in an image at a pixel level.

When should I use instance segmentation?

You should use instance segmentation when you want to have a more in-depth understanding of objects within an image. By annotating every individual object and outlining its shape, you create a gold standard dataset that will deliver better-performing models with less data than if you use bounding boxes.

What is the difference between object detection and instance segmentation?

The short answer is that instance segmentation uses polygons or masks that follow the outline of an object whereas object detection uses bounding boxes.

What this means practically is that there’s a noticeable difference in the quality of data as bounding boxes contain much more “noise”.

For example, think about a star-shaped object like a starfish. If you draw a bounding box around a starfish, 30-40% of that bounding box is going to contain something that is not a starfish. This 30-40% is noise and will make it more difficult for your model to “understand” what a starfish should look like. Instance segmentations have (if you do it right) 0% noise as you only annotate the object itself. When training a model, instance segmentation often gives better results with less data as it gives models better data to work on.

What is the difference between instance segmentation and semantic segmentation?

Instance segmentation can sometimes be confused with semantic segmentation. To understand the difference between them, imagine that you have three people holding hands, crossing a road. If you use semantic segmentation, you annotate these three separate people with one mask, as you don’t care about the number of objects in the image. The goal is to assign all pixels to a group and the number of objects is not important.

If you use instance segmentation, you create three different annotations - one for each person. Although instance segmentation also works on a pixel-level, here the number of objects is important.

Which one to use depends on your use case, but as soon as you want to know how many of something is in an image, instance segmentation is the way to go.

I also need to add attributes to my instances, do you support that?

Yes, we support annotation attributes.

Do you offer any quality control features to use after I am done annotating?

We do. Our manual review feature lets you view all annotations in one easy to use interface while allowing you to filter so that you can drill down in your data to find errors.

How do I export my annotations?

You can export your annotations in various formats. Read more about which export formats we support here. If you need your data in a format that is currently not supported, let us know and we will add it to Hasty as quickly as possible.

The Hasty AI model is working really well - can I export that too?

You can’t export it yourself, but we can do it for you on request. To learn more about how you can get access to our model and how we can help you fine-tune and adapt the model for your use case, contact us directly.

Get in touch

Help and support

Follow us

The Hasty logotype
Making vision AI accessible

Address

Zionkirchstrasse 73a
10119, Berlin
Germany