Computer vision platform Supervisely covers entire R&D lifecycle, including data labeling. To build accurate computer vision systems you need a lot of high-quality labeled data. Supervisely provides best-in-class data annotation tools and infrastructure to organize scalable labeling process with a few clicks.
The goal of this post is to cover most useful annotation features of Supervisely that will help you to label images and videos for your computer vision application.
Try Community Edition, it is free, available online, with no limitations.
Enterprise Edition is a self-hosted solution designed for businesses to guarantee 100% data privacy.
Say goodbye to polygonal and brush tools. One year ago we announced first version of Supervisely SmartTool — interactive instance segmentation. Supervisely SmartTool is a class agnostic neural network that performs precise segmentation of dominant object within selected area.
The key feature is that users can customize it for their specific objects. Since first release we continuously collect new labeled data and improve the accuracy of our SmartTool to cover more and more use cases out of the box.
In contrast, other annotation tools on the market compel users to use weak tools based on superpixels, watershed algorithms or something similar. Users have to adjust many parameters during annotation. It seems that we are at the beginning of 2005. Labeling companies demonstrate some toy examples but it is almost impossible to use their methods in production.
Polygonal tool allows to segment areas with vector objects. Add/Delete points to correct existing shape, add polygonal holes to segment complex structures.
Just put two dots around the object, move them around to adjust detection. “Aiming device” will help you to make it faster.
Use brush to draw homogenous and precise annotation, correct it with eraser.
Also Supervisely allows to correct (fill and crop) raster objects with polygonal tool. One of the use cases is correcting Neural Network predictions.
Annotation platform at Supervisely has powerful tags engine. Possible use cases: assign tags to images and objects to prepare training data for classification models, use descriptions to prepare training data for OCR models, or just use them during annotation process as notes and comments.
Select tool helps to navigate over large number of objects on the image. Once you choose the object of interest, just move it around or start editing. Move operation is helpful for video annotation: label first frame and then copy object to the next one and slightly adjust objects position.
An image must have the proper brightness and contrast for easy viewing and labeling. The brightness / contrast adjustment is super helpful especially for images that are captured in poor lighting conditions or for medical images.
Let’s consider semantic segmentation for self-driving car. How to label objects on complex scenes: with or without intersection? Best practice is to annotate objects with intersection and then arrange their order. Opacity tool helps to see final segmentation in real time during annotation. Supervisely supports these features out of the box.
By using the Undo (Ctrl+Z) /Redo (Ctrl+Shift+Z) commands and the History panel, you can easily control the state of your annotation process.
To increase labeler productivity Supervisely lets you view a list of all hotkeys, and edit or create them. The Keyboard Hotkeys dialog box serves as a shortcuts editor, and includes all commands that support shortcuts.
For every project user can customize visual settings and default behavior: size of points and lines width, default objects opacity, tags location mode, enable grid mode, change zoom multiplier and so on.
Hope you will find the features described to be useful for your computer vision project. Save time, money and efforts on building custom infrastructure. Try Supervisely Community Edition for free or speak with us about an Enterprise solution for your business.
If you found this article interesting, then let’s help others too. More people will see it if you give it some 👏.