High-performing machine learning models require a lot of training data to develop and maintain, but obtaining and labeling this data from scratch can be a time-consuming and costly process. You need a large labeling and QA team already in place, build and train one yourself, or be lucky enough to find a service or freelancers with the necessary skills and subject matter know-how - none of which is always easy or possible. And while labeling itself can be time-intensive enough, polygon segmentation is among the most labor-intensive and precision-oriented of all annotation tasks, taking up to three or four times longer per task.
That’s where data labeling automation can come in handy. But what if you don’t have an existing model to generate pre-labels or ground-truth datasets you can use to train custom ones? Advanced tools based on interactive AI, such as AI-assisted or ML editor-assisted tools, can help significantly reduce the time and expertise required to label data without any initial training or pre-labeled data required.
With this in mind, Superb AI is excited to introduce Auto-Edit: your new AI-powered shortcut to pixel-perfect polygons - at no extra cost to all Team and Enterprise customers.
Create quality polygons and segmentation masks 4x faster with Auto-Edit
Auto-Edit is a new, class-agnostic AI-assisted annotation tool that empowers labelers and reviewers to draw and edit polygons in less than a second. This tool allows teams to automatically segment individual objects in images and videos, including complex and irregular shapes, and create pixel-accurate polygons by simply drawing a bounding box - improving throughput, accuracy, and consistency compared to manual annotation.
Auto-Edit compared to manual polygon segmentation using the pen tool
Auto-Edit is fully pre-trained to detect shapes and forms that stand out within the box you draw, so whether you’re labeling data for your first model or already have models in production, it can jumpstart any labeling project with virtually no upfront effort. Plus, Auto-Edit is a great tool to build ground truth datasets for further automating your labeling using a custom auto-label or even to improve and refine imported pre-labels from your existing models.
Here’s how to best use Auto-Edit to help automate your labeling workflows:
After setting up a new polygon segmentation project (image or video), or navigating to an existing project, access the annotation app and select your intended class, such as a car. In the upper bar, you can choose between the pen tool (mark vertices to draw), the brush tool (draw with a paintbrush), or our AI-assisted Auto-Edit tool.
Draw a loose bounding box around the object you wish to segment. The AI behind Auto-Edit will generate a polygon in milliseconds, meaning no queues are ever needed, unlike other forms of AI-assisted labeling.
Perfecting your polygon is as easy as clicking the (+) or (-) buttons to include or exclude a selected area. With each click, the AI will make the revisions for you just as fast as the first time.
Auto-Edit can also quickly and efficiently edit existing polygons, whether drawn manually with the pen tool or automatically labeled via auto-label or custom auto-label. At the same time, polygons created by Auto-Edit can also be edited with the pen or brush tool for those occasions when a human touch is needed.
Watch the demo below for a step-by-step tutorial, or read more in our documentation.
More easily work with and draw irregular polygons with vector brush tool
Pen style tools have traditionally been the easiest way to work with objects containing lots of straight lines and edges, but working with irregular shapes has always been a challenge. That’s why we’ve added a humble yet mighty brush tool that will allow you and your team to perform super-accurate polygon and instance segmentation tasks in less time without needing many points. The brush tool is also great when using a touch-screen or stylus, as you can just draw naturally.
And it streamlines working with auto-edited and auto-labeled data, as making both the initial ground truth annotations and post-labeled corrections is generally much easier with the brush than the pen (depending on the shape, of course). This tool supplements our existing pen tool and new Auto-Edit tool, so you can use whichever best fits your task or mix and match as needed.
With the new brush tool, you can:
Draw polygons by dragging the brush along your object or over it to fill it in
Use a fill function to fill a hole inside your polygon automatically
Adjust the size of your brush by pixel count
Add additional polygons or expand existing ones (union)
Remove or reduce existing polygons (subtract)
And, just as before, you can still perform functions like moving, copying, or deleting polygons, or use the pen tool to work directly with points. Plus, utilize keyboard shortcuts to move between all these functions seamlessly, meaning faster and easier-to-create polygon shapes with pixel-level accuracy.
Watch the demo below for a step-by-step tutorial, or you learn more in our documentation.
The right tool for the right task
With this release, teams now have access to a pen, brush, and an AI-assisted annotation tool when working with polygons and segmentation masks. But how to know which tool is right for your specific task? Here’s what we suggest:
For objects that are clearly distinguishable from the background, start annotating an object using Auto-Edit, and finish it off (if needed) using a combination of the pen and/or the brush tool.
Use the pen tool to label small, blurry, or heavily occluded objects manually.
Once you have enough labeled data, train an auto-label model, apply it and edit the results using the pen and/or the brush tool.
Get started today
Auto-Edit for images and video (sequence of images) is currently available to all Team and Enterprise customers, and the brush tool is available in all plans. We’re excited for our users to experience the power of automation that combines a human touch with AI precision and speed, and to experience how it can dramatically cut labeling costs while improving throughput.
Try our Auto-Edit tool on a labeling project now, or get in touch with our team for help with any projects or questions you have.
