Speeding up your labeling work is what Segments.ai is all about. Today we’re excited to launch a new feature that helps you segment images even faster.
Creating image segmentation labels is already very easy on Segments.ai: with our Superpixel tool you can label an entire image in just a couple of clicks. The Superpixel tool is however not ideal for labeling thin structures and very high-res images. This is where Autosegment comes in.
Using Autosegment is as simple as drawing a box around the object you want to label. The tool then automatically generates a segmentation mask for the foreground element in the box. Autosegment can handle a wide variety of shapes and is designed to deal with imprecise bounding boxes. This means you don’t have to draw a perfect box in order to obtain a perfect segmentation mask. Additionally, you can also use the tool in eraser mode to remove (parts of) an object from the active selection.
Autosegment is a powerful companion to our superpixel tool. Try it out and let us know what you think!
As you might have noticed, we’ve recently made some cosmetic changes to our platform to give it an overall fresher look. This is part of a larger redesign project where we will improve the UX/UI of some pages, so don’t be surprised if you see some more changes in the coming weeks.
We’ve optimized the performance of the point cloud cuboid interface for scenes with many cuboids. Whereas previously, the interface could become slow and laggy as the number of cuboid annotations increased, it now stays running smoothly even on older hardware.
Other features and improvements
In the image labeling interfaces, hovering over an object-level attribute now colors all objects by their value for that attribute. This makes it easy to visually verify if all object-level attributes are set correctly.
In the image labeling interfaces, you can now bring the keyboard focus to the image-level attributes by pressing hotkey “x”.
You can now view the sample and label details in JSON format from within the labeling interface, by clicking the info icon that appears when hovering over the sample name.
When a labeler or reviewer is inactive for longer than 5 minutes, this idle time is no longer counted in the labeler metrics on the overview page but recorded separately.
In the point cloud cuboids interface, you can now choose to keep the dimensions of the cuboids locked across the sequence.
Improved performance and bug fixes in the point cloud interfaces.
In the Python SDK, image sequence datasets can now be exported to YOLO format as individual frames.