OneVision Updates

New features

• Target device saved with each version — The target device and version are now stored with each toolchain version instead of only at the project level. Restoring an earlier version also restores its target device settings.
• Upper and lower thresholds for Titan Anomaly Detector — The Titan model now supports both an upper and a lower threshold.
• Mismatch filter for the Classifier tool — Classifier results can now be filtered to show only mismatches.

Bug fixes

This release addresses a number of issues reported by users:

• Heatmap in Vega mode — Fixed the Classifier heatmap not working with the Vega model.
• OCR label angles — Fixed OCR labels gaining angles after an upstream tool or ROI was reprocessed.
• Class list overflow — Fixed the class list overflowing its panel.
• SAM label loss — The last unapplied label is no longer lost when going straight to training.
• Titan model description — Corrected the description shown for the Titan model.

New features

• User trainable OCR — Fine tuning support is now available.
• Image download — Images can now be downloaded from a project up to 10 at a time (from 'Input Image')
• Default role assignment — Role assignment for labeled images now defaults to "Only images with no assigned roles"

Bug fixes

This release contains minor bug fixes for the following:

• Runtime Toolchain export
• Segmenter multi-image selection on the Process page
• Workspace path display for moved projects
• Project archive import

Bug fixes

This release contains minor bug fixes for the following features:

• Segmenter labels
• OCR statistics
• ROI when using ROI-per-blob

New features

• "Get Support" link — A new "Get Support" link has been added under ‘?’ menu for quick access to support resources, including how to create a support case.
• Support file export — A new option in Project Settings allows users to generate a support file to assist with troubleshooting when creating a support case.
• Copy/Paste Model — Users can now copy a model directly within the tool setting, to a new tool of the same type.
• Right-click menu reordering — The order of features in the tool's right-click menu has been updated for improved usability.
• Unsaved AI-assisted label warning — A confirmation dialog now appears when navigating away from an image while an AI-assisted label edit is in progress.
• Workspaces link in toolbar — A Workspaces shortcut has been added to the top toolbar, after the ‘project’ link for faster navigation.
• Absolute mode for Locator – Nova model — The Locator tool in Nova mode now offers an ‘Absolute’ option. Absolute reports the feature's absolute, global angle from -180 to 180°; Useful for fixturing.
• Deployment target downgrade prevention — Users are now prevented from changing a project's deployment target to an earlier version.
• Changed maximum detections limit for Nova Locator models — When training a Nova Locator model for a counting application (given a dataset with +5 labeled boxes per sample), we now lift the internal default limit of 100 to unlimited predicted detections.

Bug fixes

This release addresses a number of issues reported by users:

• Cached image error on project name input — Resolved an issue where an error message would appear repeatedly while typing a project name.
• ROI resize/change blocked when filter is active — Users can now resize or reposition the region of interest even when a filter is applied in the left panel.
• Sorting parameters not working — Several result sorting options that were not functioning correctly have been fixed.
• Project management screen failing to load — Fixed an issue that prevented the Project Management screen with multiple rows of projects, from loading.
• Error when adding tags to multiple views — Fixed errors when adding tags to multiple views at once.
• Anomaly Detector ROI invalidation — The Anomaly Detector now correctly invalidates results when the region of interest is changed.

New features

• When creating a new project, the default deployment target is now pre-selected to 26.1 / 3900 instead of requiring manual selection.
• The "Scaled" checkbox has been removed from the Locator- Nova Model. The tool now always operates in scaled mode.
• Smart document search is now enabled and visible in the application menu.
• The project management info panel now shows the deployment target associated with each project.
• Additional refinements in new model naming Training model descriptions.
• Two new toggle options have been added to the OCR tool's Graphics flyout menu: "Show character class names" (Alt+N) and "Show Fielding" (Alt+F). Both default to enabled.

Bug fixes

• Fixed a bug where typing in a tool parameter text field caused the input to lose focus on every keystroke. Also fixed toggle parameters (e.g. Mask From Blob, ROI Per Blob) being incorrectly reset to their default "on" state.
• Fixed a bug where applying an AI label in the Segmenter incorrectly switched the active labeling tool to the pencil tool.
• Fixed a bug where the image count shown in the project management panel was not updated after images were added or removed.
• Fixed version export logic when using beta (Export Runtime Toolchain)“ functionality. The correct (latest) runtime toolchain version is now exported instead of always returning version 1.
• Fixed a bug where switching the Deployment Target caused the Target Device Version to be reset to its default value. Added warning message if the new Deployment Target cannot set to the old Target Device Version.
• Fixed a bug where the "Simple Region" parameter checkbox disappeared from the Segmenter tool's processing step.
• Fixed a bug where the "Mask From Blob" and "ROI Per Blob" toggles could not be turned off in a tool chain configuration after completing the Apply ROI step.
• Multiple Bugfixes and small improvements in the OCR tool training.

New features

• Model name updates: The model names of all tools have been updated. An information panel with model details and guide has been added when users hover over model options.
• Project export/import and cloning rework: A new code-based mechanism replaces the previous approach for importing and exporting projects. To share a project, the user can generate an export code from the project settings; the recipient can use it to import the project into the same or a different tenant without sharing a full archive. The code and its associated link are valid for 24 hours. Project cloning now uses this same mechanism.
• E-mail address visibility changes: Non-admin users in a tenant can only see users within the same email address domain.
• All tools can be trained without a test image: Some tools and models previously required at least one labeled image with the test role in order to train. This limitation is removed to simplify the initial training. It is still highly recommended to add labeled images to the test role in order to fully validate the tool performance before deployment.
• Better error message for multi-selection: A new warning message is displayed when the manual multi-select (for bootstrapping, mass tagging, etc.) reaches the image limit. This allows users to use manual-multi select more effectively.

Bug fixes

• Fixed a display problem where every thumbnail appeared purple with "train" label after "assign role to no assigned images"
• Fixed a bug in which the selected view count didn't match the view count in the histogram bar when filtering in cumulative score graphs.
• Fixed the "error loading from cache" issue.
• Fixed various issues with Project versioning.
• Fixed a bug in which Multi-ROI colors did not match between labelling pane and main pane.