Releases · OpenSportsLab/VideoAnnotationTool

12 Apr 12:36

v1.2.0

a1b7ba6

v1.2.0 Latest

Latest

Improve dataset state management, undo/redo behavior, project close flows, and annotation workflows across classification, localization, description, and dense description. Add broader GUI regression coverage and polish media and editor interactions.

What's new

Centralized dataset runtime state around the Dataset Explorer flow
Improved undo/redo behavior for dataset and annotation mutations
Better dataset header inspector editing and JSON preview handling
Safer unsaved-changes workflow when closing projects
Immediate manual-save behavior in Classification
Autosave support in Description
Improved Localization smart annotation, head/label management, and label color handling
Chronological ordering guarantees for Dense Description and Localization events after add/edit/delete
Updated media controls, including mute support
Broader GUI regression coverage across lifecycle, explorer, media, history, and mode workflows

Fixes and polish

Fixed dense description removal behavior
Improved tab visibility and tab-removal behavior in editor panels
Refined controller boundaries and signal-based coordination across modules
Updated tests and workflow coverage for recent projects, dataset explorer behavior, and mode-specific interactions

Commits:

Merge pull request #11 from OpenSportsLab/dev-silvio
Fix dense description test: use on_media_position_changed instead of mocking player.position
Copilot advise: rename question in dialog
feat(localization): Adjust button styling and hide tab bar when only one tab remains
feat(localization): Implement label color management and UI integration
feat(classification): Implement head management with add, rename, and delete functionalities
feat(classification): Hide training tab and refactor tab removal logic
feat(localization): Enhance localization inference with head context and configuration management
feat(localization): Enhance smart event handling and UI interactions
Refactor Localization and Classification Controllers
fix: dense desription removal was not working
feat: Ensure chronological order of dense and localization events after add, modify, and delete operations
feat: Enhance Dense Description module with improved signal handling and controller API
feat: Enhance Description module with improved signal handling and API methods
feat: Improve column layout handling in DenseAnnotationPanel with retry mechanism
feat: Add mute button with icon and tooltip to MediaCenterPanel
Refactor tests to use dataset_explorer_controller directly
feat: Refactor media handling by removing QMediaPlayer dependencies and implementing signal-based playback control
feat: Enhance dense description functionality with new event handling and UI updates
feat: Enable immediate saving of manual annotations and update confirm button visibility in ClassificationAnnotationPanel

What's Changed

Refactor dataset state management and stabilize annotation workflows across all modes by @SilvioGiancola in #11

Full Changelog: v1.1.6.1...v1.2.0

Contributors

SilvioGiancola

Assets 5

07 Apr 18:25

SilvioGiancola

v1.1.6.1

902e5f1

v1.1.6.1

Release Notes

Scope

This release includes all changes merged from:

PR #8 (fd273b4)
PR #9 (cff7c50)
PR #10 (902e5f1)

Highlights

Major UI/controller architecture simplification across all modes.
Full migration to Qt Designer .ui-driven panels for core app surfaces.
Dataset Explorer is now the central owner of sample lifecycle and selection routing.
Persistent Recent Datasets in Welcome screen (QSettings-backed).
Large pytest-qt GUI suite expansion for lifecycle and mode workflows.

New and Improved

Unified and simplified app structure

Flattened ui/ structure (removed deep nesting and ui/common).
Flattened shared controllers (removed controllers/common).
Standardized package pattern per panel: __init__.py + .ui + README.md.

Dataset Explorer-first workflow

Terminology normalized from “video/clip” to sample in explorer flows/UI.
Add/remove/filter/clear operations centralized in DatasetExplorerController.
Selection contract now uses canonical Data ID dispatch to mode controllers.
Multi-view media routing handled centrally through Dataset Explorer + media player.

Navigation and media routing cleanup

Media selection is routed directly from Dataset Explorer to central media player.
Removed dead/duplicate navigation code from mode controllers.
Navigation controls were relocated to mode-appropriate widgets.
Filter behavior improved: when a filter leaves no visible samples, media and annotations are cleared.

Mode panel/controller refactors

Description and Dense Description moved to single-controller architecture with .ui panels.
Localization moved to .ui panel + consolidated controller architecture.
Classification panel and controller were aligned with the same architecture direction.
Old/unused child widget files and obsolete manager patterns were removed.

Welcome and project lifecycle UX

Welcome screen now has Recent Datasets integration with persistent storage via QSettings.
Improved close workflow for dirty projects with explicit options:
- Save
- Save As
- Close Without Saving
- Cancel
No unsaved-changes popup when dataset is clean.

Testing and Quality

Expanded GUI coverage with pytest-qt across:
- app lifecycle
- recent datasets
- dataset editing
- classification/localization/description/dense workflows
Added deterministic test fixtures, headless defaults, and QSettings isolation.
Added tests for routing and filter behavior (including empty-visible-list clearing behavior).

Breaking/Internal Migration Notes

Import paths changed after UI/controllers flattening (clean-cut, no compatibility aliases).
DatasetExplorerController now owns add/remove/filter/clear logic; mode controllers no longer own these flows.
Selection wiring is now Data ID-based (on_data_selected(data_id) style).
Some older media/player navigation signals and legacy dispatcher paths were removed.
If you maintain local extensions/plugins, update imports and controller entrypoints accordingly.

Notes

“Show Smart Labelled” remains future-ready for Description/Dense; current behavior may produce no smart matches for those modes until smart labeling is introduced there.

Assets 5

05 Apr 14:36

woshimajintao

V1.1.6

8380e24

VideoAnnotationTool_V1.1.6

Data Sandbox Logic: Implemented strict data isolation between Manual and Smart annotations, featuring a dual-list UI (Predicted vs. Confirmed) for accumulated AI results.
OSL JSON Standard: Standardized export format by merging all events into a single array, using confidence scores to distinguish AI results and ensuring strict integer types for timestamps.
Robust Undo/Redo: Added comprehensive history support for Smart Annotation (Run, Confirm, Delete) with tab-aware UI refreshing to prevent view-switching bugs.
UX & Stability: Added an indeterminate progress bar for inference, fixed schema registration for new projects, and implemented full state resetting when switching or creating projects.
Fix some bugs of deployment(ci.yml and release.yml)

Assets 5

31 Mar 06:07

woshimajintao

V1.1.5

e22b980

VideoAnnotationToolV1.1.5

Introduced a compact dynamic label flow layout that replaces the rigid 2-column grid, with auto-sized buttons and natural wrapping for better space usage.
Added quick timestamp editing with a Set to Current Video Time action for fast event updates.
Improved event list navigation so newly created or modified events remain highlighted and auto-scroll into view after re-sorting.
Refined the interface with reduced button height and a cleaner blue + Add Label style to minimize visual clutter.
Integrated the training workflow with soccernetpro through a background QThread, keeping the UI responsive during training.
Added real-time training monitoring with a progress bar and status updates by parsing live stdout/stderr logs.
Fixed configuration and path issues, including missing keys and external-drive dataset support.
Added user controls for safe training interruption and automatic checkpoint saving next to annotation files.
Documented FFmpeg-based fixes for macOS video decoding issues with unsupported formats.

Assets 5

11 Mar 01:13

woshimajintao

V1.1.4.1

bc1b7e0

VideoAnnotationTool V1.1.4.1

Modified apply_action_filter to make Hand Labelled and Smart Labelled filters inclusive rather than mutually exclusive.
Added real-time filter refreshing and refactored logic for batch confirmation, undo/redo, and robust confidence parsing.
Added logic to reset smart annotation UI components in the right panel within the _clear_workspace method.
Inserted filter refresh calls in refresh_ui_after_undo_redo to ensure the left list syncs immediately after undo/redo.
Fixed key mismatch in batch undo logic by changing old_batch to old_data to match the history stack.
Corrected undo key names and updated batch results to pass rich dictionaries with confidence instead of plain strings.
Added descriptive "Start:" and "End:" labels before the batch inference selection comboboxes in the UI.
Fix Save/Export button state to properly recognize confirmed smart annotations in Classification mode.

Assets 5

09 Mar 12:23

woshimajintao

V1.1.4

eefb3fb

VideoAnnotationTool V1.1.4

Implemented dynamic label mapping in config.yaml to remove hardcoded class dependencies.
Fixed cross-loading bugs; eliminated the need to start from an empty JSON; fixed task-name validation errors.
Implemented dual-track export (Hand/Smart labels); refined schema to ensure standard action_classification output.
Implemented "hard-root" logic; eliminated redundant dataset copying by referencing source paths directly.
Introduced a compact tabbed interface (Hand/Smart); optimized vertical spacing and implemented dynamic font scaling.
Developed independent Smart Annotation confirmation; added a 4-state filter (All/Hand/Smart/None).
Expanded the Undo/Redo system to support batch operations and AI inference tasks.