labelimgplusplus 3.2.0__tar.gz → 3.4.0__tar.gz

{labelimgplusplus-3.2.0 → labelimgplusplus-3.4.0}/HISTORY.rst

@@ -1,6 +1,135 @@
 History
 =======
 
+3.4.0 (2026-08-12)
+------------------
+
+Portable whole-video propagation release adding multi-object OpenCV tracking,
+durable gap records, atomic preview and commit, and bounded correction
+regeneration while preserving legacy pending observations, annotation formats,
+filenames, command-line entry points, plugin APIs, shortcut identifiers, and
+Python 3.8 through 3.13 support. Optional SAM 2 propagation remains deferred to
+3.5 and is not bundled or downloaded by this release.
+
+Propagation and Gap Contracts
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* Add frozen, Qt-free propagation requests, batches, results, and gap records,
+  plus a stateless backend interface. Workers exchange immutable plain data;
+  Qt widgets, canvas shapes, model mutation, and undo commands stay on the GUI
+  thread.
+* Upgrade video projects transactionally from schema v1 to v2 with inclusive,
+  revisioned ``track_gaps``. Writable migration failure rolls back and reopens
+  read-only with an actionable warning, while read-only v1 projects retain an
+  empty gap projection.
+* Preserve accepted, pending, and rejected legacy observations and extend
+  snapshots and save deltas for gap insertion, replacement, deletion, undo,
+  restoration, export, and revision conflicts.
+
+Portable Whole-Video Propagation
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* Add **Propagate across video** and **Propagate selected object** beside the
+  integrated timeline. Existing directional commands now use the same accepted
+  atomic result path rather than creating new pending suggestions.
+* Decode each requested direction once and update all active tracks together
+  with bounded Lucas-Kanade flow and affine estimation. Rectangles transform
+  their corners, polygons transform every vertex, and associated keypoints use
+  the same affine transform.
+* Protect every accepted manual anchor. Reaching another manual anchor retains
+  and reseeds it; a failed track records a stable ``low_confidence``,
+  ``occluded``, ``out_of_frame``, or ``scene_cut`` gap while other tracks
+  continue.
+* Stream generated observations into a preview-only canvas layer excluded from
+  canonical shapes, dirty state, saves, exports, plugins, inspectors, and undo.
+  Frame navigation remains available while geometry and class mutation, save,
+  export, undo, and redo are disabled.
+* Validate document generation and revision, media fingerprint, stream and time
+  base, and every track revision before applying accepted tracker observations
+  and gaps in one model revision and one undo command. Cancellation, stale
+  state, source replacement, shutdown, or decoder failure clears preview state
+  without changing the model or durable project.
+
+Correction Regeneration
+~~~~~~~~~~~~~~~~~~~~~~~
+
+* Editing tracker or interpolated geometry immediately creates an accepted
+  manual anchor as its own undoable correction.
+* Regenerate only frames strictly between that correction and the nearest
+  manual anchor on each side, using media bounds where an anchor is absent.
+  Successful replacement affects only generated observations and overlapping
+  gap portions for that track and those intervals, and is recorded as a second
+  undo entry.
+* Cancel superseded regeneration and discard results after correction undo,
+  intervening document or track revision, document replacement, or shutdown.
+  Failures preserve the correction, all manual anchors, other tracks, and prior
+  generated data.
+
+Qualification
+~~~~~~~~~~~~~
+
+* Cover CFR and VFR media, forward and backward propagation, rectangles,
+  polygons, keypoints, multiple tracks, manual-anchor reseeding, scene cuts,
+  occlusion, out-of-frame loss, corrupted media, read-only and legacy projects,
+  cancellation, stale revisions, atomic undo, gap persistence, save/export
+  blocking, navigation during work, document close, and repeated teardown.
+* Qualify Python 3.8 through 3.13, plugin, SAM, video, and combined extras while
+  keeping PyAV, NumPy, OpenCV, ONNX Runtime, Torch, and SAM 2 lazy or absent from
+  the base import path as applicable.
+* Review the integrated propagation preview at fixed Linux 1366x768 and
+  1440x900 approval sizes with no clipped primary controls, docks, or detached
+  windows.
+
+3.3.0 (2026-08-12)
+------------------
+
+Faster-annotation release adding non-modal class confirmation and selectable
+Smart Select geometry while preserving the 3.2 workspace, annotation formats,
+filenames, command-line entry points, plugin APIs, shortcut identifiers, and
+Python 3.8 through 3.13 support.
+
+Inline Class Confirmation
+~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* Keep completed Box, Polygon, Smart Select, and video geometry in a transient
+  provisional layer until its class is confirmed. Provisional geometry is
+  excluded from canonical canvas shapes, the Objects inspector, dirty state,
+  saves, plugin snapshots, and undo history.
+* Open a non-modal, screen-clamped class picker beside new geometry with
+  filtering, arrow-key navigation, valid new-class entry, Enter confirmation,
+  and Escape cancellation. Confirmation preserves the active tool and canvas
+  focus so repeated annotation remains uninterrupted.
+* Commit confirmed image or video geometry through one ``CreateShapeCommand``.
+  Video confirmation creates an accepted manual anchor; cancellation leaves
+  both the in-memory model and durable project unchanged.
+* Retain session-only last-class memory. Default-label mode bypasses the picker,
+  while single-class mode bypasses it only after the first class is confirmed
+  in the current session.
+
+Smart Select Output Modes
+~~~~~~~~~~~~~~~~~~~~~~~~~
+
+* Add the frozen plain-data ``SamResult(polygon, bounds)`` contract. SAM workers
+  return a simplified polygon and tight, non-zero, half-open bounds for the
+  same largest mask component without mutating Qt or document state.
+* Add a contextual Box/Polygon selector that appears only while Smart Select is
+  active. Its validated ``sam/outputMode`` setting defaults to ``polygon`` and
+  persists independently of model settings.
+* Route both output types through the provisional picker and the same image and
+  video undo path. Existing stale-generation and cancellation guards remain in
+  place, and ONNX Runtime, NumPy, OpenCV, and model loading stay lazy.
+
+Qualification
+~~~~~~~~~~~~~
+
+* Cover image and video confirmation, cancellation, focus, filtering, repeated
+  annotation, bypass modes, output bounds, simplification, persistence, stale
+  SAM results, manual video anchors, and one-step undo.
+* Qualify the picker and Smart Select selector in light and dark themes at 1x
+  and true 2x DPI using 1366x768 and 1440x900 logical approval sizes.
+* Make queued Gallery and annotation-catalog refreshes inert after window
+  shutdown so late Qt callbacks cannot submit work to a stopped coordinator.
+
 3.2.0 (2026-08-11)
 ------------------
 

{labelimgplusplus-3.2.0 → labelimgplusplus-3.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: labelimgplusplus
-Version: 3.2.0
+Version: 3.4.0
 Summary: labelImg++ is an enhanced graphical image annotation tool with Gallery Mode for browsing and labeling images
 Author-email: Abhik Sarkar <abhiksark@gmail.com>
 License: Copyright (c) <2015-Present> Tzutalin
@@ -78,7 +78,7 @@ polygons, and keypoints, designed for machine learning and computer vision
 projects. It is forked from the original LabelImg with significant
 enhancements.
 
-    **Version 3.2.0 (stable).** Install with
+    **Version 3.4.0 (stable).** Install with
     ``pip install labelimgplusplus``.
 
 .. image:: https://raw.githubusercontent.com/abhiksark/labelImg-plus-plus/c7fbd5fc08a561206b210706143a50023c82a782/resources/demo/demo.gif
@@ -132,6 +132,27 @@ Workflow and Interface
     annotation-status filter close at hand, and the inspector can be collapsed
     when maximum canvas space is needed.
 
+**Faster Class Confirmation**
+    New boxes, polygons, Smart Select results, and video geometry remain
+    provisional while a non-modal class picker opens beside the annotation.
+    Filter or enter a class, press **Enter** to commit one undoable annotation,
+    or press **Escape** to discard it without changing the document. Default
+    labels and established single-class sessions bypass the picker.
+
+**Smart Select Output Modes**
+    While Smart Select is active, choose **Box** or **Polygon** from the compact
+    canvas control. The choice persists between sessions, and either result
+    follows the same provisional class-confirmation and undo workflow.
+
+**Portable Whole-Video Propagation**
+    From the integrated timeline, propagate every accepted manual anchor on
+    the current frame or only the selected object. The portable OpenCV backend
+    processes rectangles, polygons, and associated keypoints together, shows
+    preview-only progress, protects later manual anchors, and commits accepted
+    observations and explicit gap records atomically in one undo step. Editing
+    generated geometry creates a manual correction first, then regenerates only
+    the bounded neighboring segments as a separate undoable change.
+
 **Direct Ultralytics Dataset Export**
     Choose **Tools → Export Ultralytics Dataset…** to create a ready-to-train
     YOLO detection dataset with ``images/{train,val,test}``, matching
@@ -401,7 +422,7 @@ Or use **Menu > File > Reset All**
 Release History
 ---------------
 
-This source tree identifies itself as **3.2.0**, the stable 3.2 release. See
+This source tree identifies itself as **3.4.0**, the stable 3.4 release. See
 the `release history
 <https://github.com/abhiksark/labelImg-plus-plus/blob/master/HISTORY.rst>`__ for
 version-by-version changes and `GitHub Releases

{labelimgplusplus-3.2.0 → labelimgplusplus-3.4.0}/README.rst

@@ -26,7 +26,7 @@ polygons, and keypoints, designed for machine learning and computer vision
 projects. It is forked from the original LabelImg with significant
 enhancements.
 
-    **Version 3.2.0 (stable).** Install with
+    **Version 3.4.0 (stable).** Install with
     ``pip install labelimgplusplus``.
 
 .. image:: https://raw.githubusercontent.com/abhiksark/labelImg-plus-plus/c7fbd5fc08a561206b210706143a50023c82a782/resources/demo/demo.gif
@@ -80,6 +80,27 @@ Workflow and Interface
     annotation-status filter close at hand, and the inspector can be collapsed
     when maximum canvas space is needed.
 
+**Faster Class Confirmation**
+    New boxes, polygons, Smart Select results, and video geometry remain
+    provisional while a non-modal class picker opens beside the annotation.
+    Filter or enter a class, press **Enter** to commit one undoable annotation,
+    or press **Escape** to discard it without changing the document. Default
+    labels and established single-class sessions bypass the picker.
+
+**Smart Select Output Modes**
+    While Smart Select is active, choose **Box** or **Polygon** from the compact
+    canvas control. The choice persists between sessions, and either result
+    follows the same provisional class-confirmation and undo workflow.
+
+**Portable Whole-Video Propagation**
+    From the integrated timeline, propagate every accepted manual anchor on
+    the current frame or only the selected object. The portable OpenCV backend
+    processes rectangles, polygons, and associated keypoints together, shows
+    preview-only progress, protects later manual anchors, and commits accepted
+    observations and explicit gap records atomically in one undo step. Editing
+    generated geometry creates a manual correction first, then regenerates only
+    the bounded neighboring segments as a separate undoable change.
+
 **Direct Ultralytics Dataset Export**
     Choose **Tools → Export Ultralytics Dataset…** to create a ready-to-train
     YOLO detection dataset with ``images/{train,val,test}``, matching
@@ -349,7 +370,7 @@ Or use **Menu > File > Reset All**
 Release History
 ---------------
 
-This source tree identifies itself as **3.2.0**, the stable 3.2 release. See
+This source tree identifies itself as **3.4.0**, the stable 3.4 release. See
 the `release history
 <https://github.com/abhiksark/labelImg-plus-plus/blob/master/HISTORY.rst>`__ for
 version-by-version changes and `GitHub Releases