Changelog

May 2021 Update

May 5, 2021
A series of cards from V7 Darwin

Scan Text Automatically 

Your team will never have to manually key in text again. Use the new Text Scanner model to detect and scan the text in any of your image datasets. 

A Better Brush Tool

Any polygon being adjusted with the Brush Tool now has a dashed appearance - Making it stand out on the canvas.
You can also enter the exact pixel size needed when adjusting the the Brush Tool's tip size.

A Richer Sidebar


The Annotations menu now displays each label's author and lists all sub-annotationsthat have been added.
Text, attributes, and other sub-annotations can be added directly from the sidebar.

New Filter: Comments


Quickly filter for all files that have received feedback from your team.

Drag to Rerun Auto-Annotate


Dragging the edges of the Auto-Annotate bounding box will now trigger the modelto rerun to track objects in video.

Reset Workflow in Bulk


Send files back to a New state to restart workflow without losing any of your annotations.

More Features and Improvements

Feature
Feature
Feature
Feature
Performance
Performance
Performance
Performance
Fix
Fix
Fix
Fix
Fix
Fix
Fix


New Features  

🆕 PDF support.
🆕 DICOM DIR Support.
🆕 Auto-stopping for models
🆕 The video timeline now jumps to any frames selected outside the current view.
🆕 No need to name a new workflow before saving.
🆕 Datasets are now filtered by Newest by default, and sort order is remembered.
🆕 New confirmation dialog to prevent unsaved changes in Dataset Settings.


Performance Improvements  


🚀 Removed unused counts to increase performance.
🚀 Sped up queries when filtering by assignee. 
🚀 Importing annotations is greatly sped up; more annotations can be handled at once.

 Fixes  

🛠 Fixed bug preventing copy & paste in videos.
🛠 Fixed errors in open datasets.
🛠 Fixed bug preventing exports from some datasets.
🛠 Fixed how model stage annotations are inserted.
🛠 Fixed the model instance count for daily/monthly/total aggregations.
🛠 Fixed issue with filtering annotation classes for exports.
🛠 Fixed issue with restoring image endpoint.
🛠Fixed response error on class delete endpoint.
🛠 Fixed issue with credit transfer during upgrade.
🛠 Fixed jumping in frame when adjusting class length.
🛠 Setting images to new now triggers auto-start for model stages.
🛠 Datasets with underscores in their names now download correctly.
🛠 Fixed a bug which prevented datasets from loading right awayafter accepting invitations.

April 2021 Update

April 6, 2021
A series of cards from V7 Darwin

Search by Filename, Folder Name, and Annotator 

You can put Ctrl + F away for good.  Search for any item within a dataset in the Data page. No matter the size, you'll be able to retrieve any file within million-image datasets.

TIP Shift + Click  to add multiple options in your search query.

A New Report: File-Level Annotator Statistics


Download the Overview Report from the Dataset Overview page to see a breakdown of all of your images and videos, when were they added, when did labelling start, who worked on them, and for how long.

Train Neural Networks on Video Frames


You can now train models on videos whose frames are imported as individual images.

Measure Regions for Images, Video, Ultrasound, and DICOM!


Annotations now contain either pixel or millimeter measures. Click on the new Measurements button while annotating to switch these on.

Copy Filenames With a Click


Click on any item's filename to save it to your clipboard.

Reset Items to new



If you ever need to start from scratch, select all and bring entire datasets back to brand new.

Heads up! This will remove all existing annotations.

More Features and Improvements

Feature
Feature
Feature
Feature
Performance
Performance
Performance
Performance
Fix
Fix
Fix
Fix
Fix
Fix
Fix

New Features  

🆕   Generate exports based on filters.


🆕   Use partial filenames when searching for items from the API.


🆕   Set up workflows to automatically start labelling items with a Model stage.


🆕   Filter by assignee for annotation tasks in previous workflows.


🆕   Quickly set up workflow stages with only team members with User permissions and above using the Any User filter.


🆕   Keep unused credits when upgrading usage tiers.


🆕  Creating a new team is now even easier.


 Performance Improvements  


🚀   Processing speeds for video after you upload are now 46x faster.


🚀   Select All load times reduced to almost nothing for large datasets.


🚀   Faster annotator filters.


🚀   Increase in size limit (8mb to 32mb) for images uploaded to the model inference section of a trained Neural Network.


 Fixes  


🛠   Fix to reject bounding boxes with a negative width or height.


🛠   Bounding boxes render correctly on model inference page for Neural Networks.


🛠   Improved auto-Annotate performance on tiled images like SVS or ultra-high res.


🛠   Improved rendering of bounding boxes after mouse release.


🛠   Several safari-related display improvements.


🛠   Fix zoom to focus on annotation when Tab is selected.


🛠   Fix to allow for the addition of points to polygons and polylines.


🛠   Annotation centroids no longer take mouse control when creating annotations.


🛠   Model stages no longer remain stuck if the model is offline.


🛠   Multiple improvements to video playback in workview.


🛠   Fix for an issue where keypoint skeletons did not appear in the Complete stage.


🛠   Fix ordering of image carousel when items are moved to a new stage.


🛠   Fix to prevent invalid annotations from crashing exports.


🛠   Fix to prevent MONOCHROME1 DICOM images from inverting.

February 2021 Update

February 25, 2021
A series of cards from V7 Darwin

Everything is in Real-time

We want your data to feel alive. Statuses, assignments, quantities, datasets, now all update in real-time across your team.

This gives us the chance to bring even more collaborative features to V7, both for data scientists and annotators. Watch this space :)

15 times


That's how much faster annotations are now rendered on V7 compared to the sector leading web library.

500,000 points are tracked in this GIF without lag. And we wont' stop here!

No Loading


We're not just removing the need for refreshing - waiting is going as well. Load times are now instant on most file types. Over time, we'll reduce most load times to zero to offer the smoothest experience of any ML platform on the web.

DICOM is smart(er)


Just drop your DICOM studies, and V7 will re-order CT and MRI slices based on header values. Gone are the days of pre-processing your data.
Thanks to all our medical imaging users for the continuous tips and feedback!

Fun fact: That's our CEO's brain. Technically where V7 was born.

Video Annotation Track Size


A little window above video timelines allows you to define the default length of a new annotation. This should help reduce time spend lengthening and shortening their sizes.

TIP: You can turn any number of images into a sequence (which behaves like a video) by zipping it and renaming .zip to .mp4. Use this for multi-page PDFs, multi-camera setups, or to turn images back into a video.

Balance your classes


The dataset overview page got some new tools. Class balance histograms now show data by image and instances, and point out over/under representation of your classes. These are also isolated by annotation type.

More Features and Improvements

Feature
Feature
Feature
Feature
Performance
Performance
Performance
Performance
Fix
Fix
Fix
Fix
Fix
Fix
Fix

New Features:

:new:

Teams with high security requirements can now prevent the sharing of datasets.


:new:

Export formats now explain what each will support.


:new:

The home page received a few design improvements - from the tutorial window popups to the look & feel of buttons, richer dropdowns, and modals.


:new:

Split videos can now be exported without having to select individual frames


:new:

Support for .svs uploads via the CLI


:new:

COCO exports can now be rounded to integers


:new:

Support for MONOCHROME1 DCM with Modality OP

 Performance Improvements  

:rocket:

 Videos load A LOT faster now


:rocket:

 Greatly increased the stability of running models


:rocket:

 The amount of memory required for browser uploads has been greatly reduced.


:rocket:

 Annotation imports are now 2x faster


:rocket:

 Sub annotations render more quickly when they reach large numbers.


:rocket:

 Exporting images has been sped up further and is now instant in most cases.


:rocket:

 New metrics and improvements on the class balance chart in dataset overviews


:rocket:

 FPS sampling in externally-hosted videos is now live.


:rocket:

 Settings is more clearly visible in the sidebar and now accessible via the team-switch dropdown.


:rocket:

 Deleting large images now has a more reasonable load time


:rocket:

 Subscription upgrades now reset billing periods, so that users don’t have to wait for a new period to upgrade.


 Fixes  

  •  Fix status changes deselecting selection
  •  Fix spacing issues in dataset management context menu
  •  Broken URLs in certain video exports have been fixed
  •  Fixed issues in exporting tags for videos
  •  Review stages are no longer ‘read only’ by default
  •  Images in instructions no longer break after an extended period of time.
  •  Hidden joints in occluded skeletons are now visible!
  •  Render skeleton in read-only Review mode
  •  Fixed issues with sub annotations not moving in videos
  •  Fixed issues with workflow stage name losing focus on each new input
  •  Fixes frame jumping back to the start when videos aren’t fully loaded
  •  Fix blank screen after first creating a team
  •  Fix infinite retries of failed model stage requests
  •  Fix incorrect stage type for model stages
  •  Allow archiving of dataset items with open comment threads in them
  •  Restarting workflow no longer shift z-indexes
  •  Skipped images in Archive -> Complete will now archive properly

January 2021 Update

January 27, 2021
A series of cards from V7 Darwin

Auto Annotate 5

We're on the 5th version of Auto-Annotate! Other than improving in overall performance, it can now be tweaked to create less-fine outputs.

That's right. You can make Auto-Annotate look as if it's polygon was made by a human.

Select a value of fine-ness between 1 and 5 and start labelling! Note that less-fine polygons are also more performant, in case you are labelling long videos with many polygons.

Merge and Subtract Polygons


For the semantic segmentation crowd out there. Previously the only way to cut holes into polygons or create islands was to use the Brush tool. You can now merge & subtract any polygon from another. Press SHIFT+M or SHIFT+S  and click on your target.

Comment Mentions


Press @ to start mentioning another user in your comments. They will be notified and start following the thread. When typing a user's name you can press TAB  or ENTER  to include them without having to click on their name.

Mass Status Changes


You'll see a new button on the bottom toolbar after selecting images. You can now change the status of multiple images at once!

Polyline Interpolation


Polylines now automatically interpolate in videos. Just like in other annotation types, keyframes will form each time you move the annotation or modify its shape.

Power-User Hotkeys


PressQ to open the class-change dialog. UP and DOWN to cycle options and ENTER to select.

Numerical hotkeys are also automatically assigned to the first-created classes in each dataset.
Picking the right class automatically switches to the correct tool for drawing it.
itches to the

You can do the same with Attributes -Shift+A to open the panel, and the arrow keys to navigate options.

Several tooltips for power-user hotkeys have been added throughout the platform.

More Features and Improvements

Feature
Feature
Feature
Feature
Performance
Performance
Performance
Performance
Fix
Fix
Fix
Fix
Fix
Fix
Fix

 New Features  

:new:

Quickly cycle through annotations duringReview orAnnotation stages usingTab


:new:

Power user hotkeys were added to switch between classes, change the class of an object, and assign attributes.


:new:

You can now change an annotation class by selecting it and changing it on the class dropdown or through hotkeys.


:new:

A new model creation UI has gone live for teams with access to it.


:new:

The numbers of images, annotations, and stages can now be seen within folders, indicating the values only for images found within the opened folder.


:new:

You can apply and remove classification tags with hotkeys.


:new:

Exports can now use the original filenames for their corresponding annotation file.


:new:

CTRL+A now allows for select-all of images in folders and datasets.


:new:

Settings now allow the same user to take part in multiple stages on the same workflow.


 Performance Improvements  

:rocket:

 Opening, viewing, and posting comments is now faster


:rocket:

 Speed up listing dataset folders (97% faster)


:rocket:

 Exports are now a lot faster. Darwin JSON still being the fastest.


:rocket:

  Deleting images now has a loading spinner.


:rocket:

  Video pre-loading behaviour has been greatly improved.


:rocket:

  Video zoom behaviour has been sped up 80%.


:rocket:

  Several tooltips were added across the platform and their design was improved.


:rocket:

  Export-related errors and incompatibilities are now more explainable. You'll know what is compatible, and what is not.


:rocket:

  New email notifications warn of approaching usage limits on credits and storage.


 Fixes  

  •  Billing months now restart upon plan upgrading and downgrading to allow upgrading mid-month without losing credits.
  • After using Auto-Annotate on an un-assigned images, the editor no longer closes after assignment occurs.
  •  Deleting the third vertex of a 3-point polygon now correctly deletes the whole polygon.
  •  Fixed a bug that caused “double rejections”
  •  The tutorial now allows the creation of new classes.
  •  DCM files (uppercase) are now correctly handled.
  •  Bounding boxes no longer present negative values in edge cases.
  •  Selecting class thumbnails is now faster and errors have been removed.

November 2020 Update

November 11, 2020
A series of cards from V7 Darwin

Workforce Managers

Assign one to a dataset and they'll manage it! Workforce Managers allow teams add external users to datasets without giving them User-level access to other information. V7's workforce partners can now add managers to individual datasets in a team, who are allowed to:
✅ invite & remove workers, browse dataset, assign images, set up new dataset, and edit settings
But are not allowed to:
❌ Permanently delete data, export raw image data, view team member contact details, and view any other dataset they haven't been added to.

Are you not the reading type? There's a video explanation below.

Colorful Tooling


Data should have color. We follow this design principle at V7 so users can spot data from other interface elements. All tools now reflect their selected class color when being used. Crosshairs are also more visible in greyscale images, and the appearance of points was improved.

Syncing Filters


Selections carry over between workview and dataset management! This means you can filter and sort by any parameter directly from workview, or set them in Dataset mgt and have them carried over to workview. As a result, many more filtering options were added directly within Workview.

Stage-Specific Filters


We're glad to see users taking full advantage of workflows! Some teams have as many as 10 annotation stages in their flows.
To facilitate this, we have added stage filters that allow you to only display images that are at a specific workflow stage.

Move Annotations, Points with your Keyboard


Try this:
Press TAB to cycle annotations.
Press § to cycle points
Use the Arrow Keys to move annotations or points
Hold Shift to move them faster.

This can help power users making bounding-boxes and keypoint skeletons place and edit their annotations without lifting their hands from the keyboard.

Priority


A much requested feature. You can now set the priority  of an image. A default 1-5 option is available (5 being highest) but you can also set any arbitrary number you wish.
Priority can be used in creative ways, both for task ranking, or as a sorting medium. You can sort by priority, and auto-assign by priority! Allowing you to define the queue of files to first be processed by workers.

More Features and Improvements

Feature
Feature
Feature
Feature
Performance
Performance
Performance
Performance
Fix
Fix
Fix
Fix
Fix
Fix
Fix

New Features  

:arrow_up_down:

The video timeline height is now adjustable! Drag its upper edge to make it taller or shorter. Frame widths are also adjustable via a slider above the timeline.

:arrows_clockwise:

Rejection changes carry forward! Or backwards technically. When a reviewer rejects an image, any changes they applied are now saved before the image is sent back to the previous annotator for them to review and learn from.

:fast_forward:

Panning has been sped up 50%

:film_frames:

Dropped frames no more! Improvements to video processing now make dropping/skipping frames on large/heavy videos impossible.

:speech_balloon:

A new notification system now notifies of comments and replies, as well as rejections and assignments.

:label:

You can now allow or prevent workers from creating new attributes via a new toggle in Dataset>Settings.

:1234:

Darwin's extracted frame files (full quality PNGs) from videos are now available for download in Darwin JSON format. This way you can train on either image stills or video files from the same export or dataset.

:eye:

DICOM OT files are now supported


Performance Improvements  

:page_with_curl:

Pagination was improved in workview to avoid harsh jumps in image orders.

:lower_left_paintbrush:

The brush tool’s performance was significantly increased. It interpolates better and requires less compute.

:film_frames:

Video performance on annotation-heavy data was greatly improved

:1234:

Image counting was sped up, which avoids Dataset loading times being long.

:floppy_disk:

The max size of JSON uploads is now larger at 32MB.


Fixes

:label:

Placing a tag on a video no longer causes timeline overlaps in workview.

:put_litter_in_its_place:

Archiving and deleting permissions were updated for users.

:label:

 Assigning un-uploaded items is no longer possible to avoid assignment bugs.

:construction_worker:

Fixed a bug where workers could see datasets they didn’t belong to if the dataset was open

:mailbox_with_no_mail:

Open datasets now show Skeletons

:person_with_blond_hair:

User avatars now display correctly in drop down menus.

Older