A fresh take and new look for our video annotation app makes labeling videos more accessible and faster than ever before. Plus, review requests and enhanced bulk operations keep label quality sky-high and make working at scale a breeze.
While December and January are some of the busiest months of the year, that hasn’t stopped our product team from delivering a bunch of exciting new features developed specifically with you and your labeling experience in mind.
Short on time? Watch our February 2022 Product Spotlight video for a run-through of all the new features in under two minutes 👆
Got a bit more time? Read on for a comprehensive look at everything we launched this month. 👇
What’s New in the Superb AI Suite?
A better way to label videos
Let’s be honest - annotating videos can be a real pain. Even a short 5-minute video can encompass >7.5k frames. Labeling automation, like our custom auto-label, can do a lot to relieve this pain, but there are always going to be times when manual labeling is needed, especially when creating ground truth for said automation. Since manually labeling videos can be the stuff of nightmares at the best of times, we’ve spent a long time reimagining our interface to make annotating videos with precision and speed more accessible. Designed to be more intuitive and easy to use, our new video annotation tool helps you transform complex scenes into accurate labels faster, and track objects across frames seamlessly - all so you can spend less time labeling and more time in production.
Our new video annotation tool is available today for our Team plan (as an add-on) and our Enterprise plan. Here’s a quick breakdown of the new UI you can look forward to when working with video data:
• Top bar: basic information and actions related to the current label
• Task panel: access configured object classes and activate the appropriate drawing mode
• Object panel: displays information about annotations on the canvas area
• Issue panel: add or review issue threads related to a specific frame or all frames
• Shortcut panel: quickly view and learn keyboard shortcuts to improve your throughput
• Adjustment panel: adjust transparency, image contrast, point size, and more
• Canvas: where you turn your raw data into labels
The bottom line - we wanted to make our video annotation app interface as simple and neat as possible and make labeling videos as painless as it can be for you. For this reason, we kept this new version in beta for a while to collect as much feedback as possible - and to make sure we reached this goal. Here are just a few changes our beta testers said they loved:
• New navigation arrows on the top make switching between video files within a project a breeze
• The label name (data key) is now visible by default in the top-left corner for quick identification
• Adding issues on individual frames make pinpointing and resolving problems easier
• New visibility modes for keypoint annotation enables you to quickly find the exact points you’re looking for: 1) visible points, 2) invisible but labeled points, and 3) invisible points
• Drawing individual keypoints without having to change the location of each point after the keypoint set is applied
• The new union function for polygon segmentation makes it easy to annotate overlapping objects
• Finally, a new subtract function for polygon segmentation reduces the time it takes to more accurately segment objects with features like holes
Ready to get started labeling videos?
Learn more about our video annotation app below:
Review requests for maintaining label quality
You’ve worked hard to collect the raw data your hungry models need, doubly so to turn that data into quality labels. The last thing you want to do is get started on the wrong foot. Or, if you’ve been at this for a while already and have models in development or production, the last thing you want to see is model degradation from preventable labeling errors. These errors, particularly of the structured kind, can significantly and negatively impact model performance regardless of if you are just getting started or well into production.
To help ensure your top-notch data becomes and remains top-notch labels, we’ve added the ability for administrators and managers to request that labels be reviewed by assigning them to specific team members. And to make working with large datasets even easier, you can use our existing comprehensive filters and search functionality to ‘slice and dice’ your dataset to find the exact labels or group of labels you want to be reviewed in seconds. This new feature allows you to:
• Request team members review specific labels, similar to assigning an image to be annotated
• Distribute labels to be reviewed equally or proportionally
• Approve or reject a label during the review process to confirm label accuracy
• Track labeler review performance and reviewer summaries through the use reports tab
Improved performance and tracking for bulk operations
We’ve also enhanced your ability to carry out bulk operations for data and labeling-related tasks, improving overall performance and speed so you can get more done faster and at scale. This includes:
• Delete data
• Assign project
• Change group name
• Assign labeler
• Unassign labeler
• Assign reviewer
• Unassign reviewer
• Consensus setting
Plus, we’ve added a nifty status bar so you can monitor the progress of your data and labeling bulk operations in real-time.
That’s it for now - make sure to check back to see what we have in store for you next month. We want our users to have the best experience possible when working with large datasets, so we’re always working on even more improvements and features - as well as an entire catalog of training and How-To videos to get you from start to success with the Superb AI suite. We’ll be sure to let you know when it launches!
Looking for a way to stay connected with everything going on in the world of computer vision and MLOps? Sign up for The Ground Truth newsletter to get resources, insights, and news straight to your inbox.