• Enhanced task confidence for AI tasks

• Super.Car: Improved edge smoothening of vehicle masks.

• Object Image Segmentation: introduced support for manual post-processing (HITL)
• Object Image Segmentation: added option for defining a background image as an replacement for redacted image background

• Object Image Segmentation: Updated AI model to yield more accurate results

• Fixed a bug that in account creation while inviting members to an organization

• Enhanced training data loading

‍

• Speed up AI job completion time by improving model ramp up strategy.

• Allow custom timeouts per AI deployment to optimize failure handling.

• Enhanced status information about training runs
• Enhanced model scaling to zero settings

• Fixed a bug that led to not fetching training stats after training is created

• Document Processing: improved usability by updating the Job Detail View showing a connector between a selected bounding box to the specific key value pair

• Fixed a bug that prevented to Edit Content in Tutorials

• Image.Redact: Enhanced Edit Output support to Image.Redact

• Fixed a usability issue: Task status UI is now consistent with Job status UI

• Image polygon based projects: updated to enhanced schema

‍

• Fixed a bug that lead examples not selected as default in onboarding workflow

• Invoice Parser: Added support for table extraction. Key-value and table extraction can be toggled on/off in the settings.
• Task level review

• Added support for reviews on task level
• Added endpoint to retrieve task worker distribution by worker type

• Image HYPE: Adding new randomisation parameter, allowing for the pairs of images to be served randomly to HITL

• Fixed a bug that led to errors when downloading of jobs filtered by state

• Fixed a bug that led to showing an empty task page

• Invoice Parser: support for key value pairs and tables

• Enhanced training status presentation on AI model view  
• Image HYPE: Added a parameter allowing the user to select to how many different humans the task will be send to.

• Fixed a bug in model deployment that led to incorrect reference of the model queue

• Image HYPE: Run your Human eYe Perceptual Evaluation (HYPE) projects.

• Document Redact: Added fallback to sans-serif when font family cannot be determined

• Document Redact: Improved document redact rendering

• Fixed a bug that allowed case sensitive e-mail addresses for worker invitations
• Fixed a bug to catch job client exception when a scheduled job response is requested

• Document Redact: Improvements the user experience in the onboarding flow

• Multiple enhancements for creating trainings and around versioning support

• Document Redact: Enabled Edit Output support
• Document Redact: Migrated data program to the new schema.
  

• Fixed a bug that led to changes in Task design were reset while editing
  

• Document support: PDF document edit output
• Document support: Allow editing/downloading redacted output.

• Fixed a bug that led to missing job "callbackUrl"

• Image Redact: Corrects bug which would undo redaction when polygons of the same kind overlapped.

‍

• Enhanced support for app training datasets.

• Fixed a bug that impacted sort and filter functionality for apps in an organization
• Fixed a bug that led to totals of archived apps count in orgs app list
  
  
  

• Get metric list from DP server if defined

• Fixed a bug that led to entity classes not showing up in task design in NER apps
• Fixed a bug that prevented from subscribing to the newsletter

‍

• Default measurer configuration for new schema DPs
• Hide 'Who should process' settings for AI apps

• Improves handling of AI output values for different formats.
• Improved error handling when Sagemaker deployments are under heavy load.

• Corrected handling of timeout issues in AIs

• Store more meaningful exceptions for predictions which exceeded their retry limit
• Support to track performance of AIs on app data over time

• Improved error handling

• Fixed a bug in autoscaling

• New `errorMessage` field for predictions which failed to complete. Allows easier debugging for model owners and visibility for data issues.

• Enhanced support for prediction monitoring

• Performance improvement for job downloads

• Added support for creating a model lineage.

• Fixed a bug that led to inconsistencies in the project overview stats

• Several performance improvements for prediction serving

• OCR projects: Added an optional validation task for the polygon drawing task. The number of times the validation tasks is given to the HIL is managed by a new parameter called polygon validators.

• Added Kubernetes orchestrator
• Added an async Sagemaker orchestrator
• Multiple schema updates

• CLI and performance improvements

• Improved auto-scaling for model predictions

• Image Redact Projects: performance improvements

• Fixed a bug in the app settings that led to no predictions were served in pre-processing and active learning for existing jobs.

• Performance improvement for delivering predictions in applications with a lot of jobs.

• Fixed a bug in custom entity recognition

• Fixed a bug that led to deployment issues due to erroneous state handling

• Image projects: download output image in job detail view of Redaction projects

• Fixed a bug that prevented to create from templates without application parameters

• Image polygon projects: allow rotation of annotation images.
  

• Enhanced worker qualification

• Enhanced filtering for jobs
• Performance improvements to retrieve job feedback
• Performance improvements to retrieve job stats

• Fixed a bug in onboarding that led to the "Next" button doesn't appear after importing JSON with more than one row

• Allow onboarding without parameters
• Improved user experience in accepting/rejecting AI model suggestions in Image Segmentation tool
• Improved usability for editing job tags
• Updated language throughout the application for improved consistency

• Enhanced autoscaling support

• Fixes Marketplace project description missing in index

• Ability to sort Marketplace index results by score

• OCR projects update: Edit Output now correctly redirects to the polygon edit view

• Improved error tracing for custom resource deployments

• Corrected button label text for adding worker group to project

• Improved error highlighting in data processing UI

• Fixed a bug in access management for submitting jobs

• AI Task Predictions: Allows assigning AI models on Task level.

• OCR projects: description of the image polygon class is properly displayed now

• Document Redact: Output PDF text is replaced with correct color and font size.

• Adding configurable post-processing for OCR projects

• Enhanced job submission

• Added features to support AI deployment and scaling

• Added new workflow for OCR Data Program

• Fixed failure to subscribe to workflow due to user api key in local deployment.

• Backend Monitoring: Improved backend monitoring and orchestration to enhance availability
  

• OCR data classification and extraction data program now supports non-RGB images

• Fixed a bug that sporadically led to an error condition when jobs where cancelled.

• App metrics are now returned as a list

• Fixed an issue that occasionally blocked model deployment when used for active learning
• Fixed an issue that led to an error when adding a model to an application.

• Fixed a bug which lead to performance degradation in Data Programs with high number of active instances.

• Added proxy support for AWS Textract in projects via Meta AI
• Added proxy support for Google Cloud Platform APIs

• Changed condition to check if job has big payload. Now only the job response is checked.

• Fixed a bug in the rendering of satellite imagery

• Added function to access files referenced in the output of a model

• Enabled task level feedback for project users

• Added performance improvement for computer vision models by supporting model accelerators
• Added support for async calls to third-party APIs to improve job latency

• Fixed an issue that impacted output rendering and task-list of versioned jobs
• Fixed an issue in the video bounding box annotation tool that blocked to select multiple objects by clicking with command key pressed
• Fixed an issue in the number field in the sub annotation pane not being reset when different object is selected

• Added ability to edit and toggle visibility of multiple annotation instances at once
• Added ability to add and view feedback on task level

• Streamlined activation of offline models when used for pre-labeling
• Task with large payload (>96kb) is uploaded to S3 bucket.

• Fixed an issue when updating project metrics

• Added support for big payload tasks

‍

• Added ability to apply custom metric for each app and select specific output fields on which the metric is calculated
• Added a view for weekly quality statistics

• Data Program Backend: Balanced CPU utilisation and better autoscaling behaviour for improved throughput.

• Extended model states: the states of deployed models now mirror the state of the underlying platform.

• Fixed visibility of selected training name when set under labeler pool restriction
• Fixed the presentation of image instance segmentation datapoints with multiple annotations so that they are rendered once

• Added support for email verification via a one-time code for new user signup

• Added job upload limit for freemium users
• Added prompt before discarding changes on add data page to prevent data loss

• New users need to verify email by code

‍

• Added job upload limit for freemium users

• OCR Extraction Data Program supports automatic entity extraction for trained entities

• Fixed secret passing during the source-to-image build process
• Fixed the conda environment creation and server process initialization during the deployment of AI models.
• Fixed the setup script run during deployment
• Fixed initialization of BaseModel during AI class creation. Now the model is initialized only when predict method is called explicitly.
• Fixed bug in orchestrator selection during deploying.

• Enhanced build process for containers

• Replaced print calls by logger calls.

• Fixed an issue in image instance segmentation data programs that led to an error on task submit

• Automatic pausing of unused model deployments: You can now enable a feature to automatically pause model deployments after a given time  without usage to reduce cost.
• Auditing: You can now track the time your models were online and also how often the models where used for predictions.

• Improved data point detail view layout

• Show call to action when project's task design doesn't match parameters supported by current data program

• Improved latency for model predictions in the case when another model is under heavy load

• Start again link during project creation now works correctly when using sample datasets
• Suggestions panel only visible when suggestions available
• Adjusted project creation process for Free Trial projects which use AI

• Enhanced error message when unsupported file types dragged and dropped

• Auto saving response no longer gets marked as completed on expiration

• Unprocessed Markdown text no longer displays momentarily in the instructions within the labeling interfaces

• Autosave your work in the labeling interface after 5 seconds of inactivity

• No longer need to choose who labels your data when the project type only uses AI for data processing

• Create and deploy models locally and remotely with Amazon SageMaker
• Version your AI models. Create copies of models whenever you adjust weights or other parameters. Every model version remains operational and accessible.
• Adjust weights and other parameters on your models
• Add models to a project
• Assign models for pre-labeling. Use a model to provide suggestions for human labelers to accept, adjust, or ignore in the labeling interface.
• Directly predict using deployed models. Use prediction to debug your model or provide a direct labeled output.
• Deploy and test models locally with new CLI tools

• Enjoy more detailed SDK log outputs thanks to Rich library implementation
• Automatically verify model responses to ensure compatibility with super.AI schema
• Shorter build time for AI containers through staged builds using a source-to-image package

• How to create and manage organizations page added

• Fixed 202 response when using `job_responses` endpoint

• Scroll between data points in tutorial content and view explanation beside selected data point

• View email addresses of workers involved in each task In the data point details view. For video bounding box keypoint projects, it's also possible to see who annotated each object in each frame.

• No trainings shown on partners page when trainings are not available
• Work queue shows status of ongoing async operations (large batches of relabelings, cancellations, etc.)
• Disqualify workers from tests
• Improved performance of data point details view for video bounding box keypoint projects
• Improved initial page load time

• You can now complete all the outputs in exclusive choice labeling interfaces, even when the options are the same between questions

• Work queue is now more consistent with backend data updates
• Human vs AI statistics now appear on the Statistics page when they're meant to

• Ouput coordinates from video bounding box keypoint, image bounding box, and image polygon labeling interfaces preserve fractional numbers

• Data points load in the work queue even when there are updates occurring on those data points
• Markdown displays correctly in the task details view

• See what type of workers (AI, human) were involved in labeling the data point in the data point details view
• See how many data points were labeled by "Human + AI" on the statistics page of a project
• Copy text input data in data point or task details view to clipboard
• Invite users as Instructors to create tutorials and tests within a project

• Project settings page loads faster

• Settings page loads without requiring a response from the metrics API endpoint

• Choose which data points you want to update when changing project settings