One of our most requested features. You can now set any annotation or review stage you wish, and edit it at any point in your project. You can now create multiple review stages, or several annotation steps that precede reviews.
You'll find these new options in the Settings tab of each dataset.
You can now upload labelled videos! Video annotations can be loaded via API. Read the documentation for it.
Video keyframes now save sub-annotations, so they do not have to be set for each new frame.
Video keyframes now support Instance IDs, so you can track and re-identify objects between video files
Original video files are now available in the export JSON, alongside the full-res frames
Instance IDs and attributes can now be applied to skeletons.
Comments are carried over across new workflows
New warning windows notify users of an expired free trial or credits plan. These windows contain links to the appropriate plans page.
An orange “External” tag is visible on images belonging to external S3 storage buckets, for teams with bespoke storage arrangements.
A new polygon rendering library speeds up the browsing of annotation-dense images. This is the first of multiple updates on faster polygon rendering.
Sped up the opening/closing of instructions, hotkeys, and image manipulation panels and reduced annoying animations.
Minor UX improvement on the plans tab when changing credits. A warning message is displayed when downgrading below a used amount.
The Annotator user class was renamed to Worker to avoid confusion (since workers can also be reviewers)
Film overlay added to video files in the workview carousel to distinguish them from images.
Fixed a bug that prevented the permanent deletion of videos.
Deleting a folder of video frames now correctly deletes all the frames within it.
We've listened to feedback, explored professional video-editing software, tested out motion-capture suites, and worked on designing the smoothest video annotation experience in the market.
When uploading a video, you will now be asked whether you want to label it either as a video, or series of frames. If you choose the former, you'll enter V7's brand new video annotation experience. Every type of label interpolates, and can be adjusted in an interactive timeline.
There's a lot that has gone into this new system, take a look at this introductory video: https://www.youtube.com/watch?v=Ik-V7PbgfPY
We couldn't fit the features developed in June in one update post.
Annotator mode has received a new design with more power-user features. Annotators can now filter images in-workview, and request a new batch of images directly from the carousel without having to exit the annotation screen. These images will start loading right away and they may continue working.
This update has been almost four months in the making. Workflows replaces the branch-based system (similar to how Git handles work-in-progress data) with a real-time experience for each image status.
You can now press ENTER to move any image to a next stage, whether it's to review, or completion. Other users can see these changes right away from the data tab, without having to wait for you to complete a whole task. This is one of the many changes Workflows brings forward. No change will disrupt how you've worked so far - you will likely notice them as improvements along the way.
We have updated the basic tutorial for V7 Darwin to reflect this major change. You can see a guide video here: https://youtu.be/v90norZUy58
Or try the actual tutorial via the button below. Make sure you are logged in to see the latest version!
and It's not just a pixel-brush - it relies on vector graphics to draw contours. This means that the output is still light just like in polygon annotations. Here's how it works:
Press F to select the brush, click and hold to draw, use [ and ] to change the brush size, and press E (or hold shift) to erase.
When you are done, press enter or switch to another tool to confirm. You can use brush to edit existing polygons by selecting them. Brush can be particularly useful for thin, elongated shapes where Auto-Annotate underperforms.