Merge branch 'master' into newpapermaster

Author: AIRCAP

DetectorNetwork/template.json

@@ -70,6 +70,23 @@ The annotations are saved as a [JSON](http://www.json.org/) file.
 ```bash
 smarter_labelme  # just open gui
 ```
+
+### Known Installation Issues.
+
+Due to a version mismatch between OpenCV and system Qt libraries, on some system the following error might occur when attempting to run smarter-labelme:
+
+```
+qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in ".../site-packages/cv2/qt/plugins" even though it was found.
+This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.
+```
+
+If this happens, you can run the following commands to remove the conflicting opencv qt libraries:
+
+```
+python3 -m pip uninstall opencv-python
+python3 -m pip install --upgrade opencv-python-headless
+```
+
 ### Neural Network weights.
 
 Smarter-labelme will automatically download pretrained network weights via torch.hub on the first start. They will be cashed in your local user directory and use approximately 200 Mb of space. You can use your own weights instead with the `--ssdmodel` and `--re3model` flags.

README.md

@@ -12,7 +12,7 @@
 # 1. MAJOR version when you make incompatible API changes;
 # 2. MINOR version when you add functionality in a backwards-compatible manner;
 # 3. PATCH version when you make backwards-compatible bug fixes.
-__version__ = '5.2.6'
+__version__ = '5.2.10'
 
 QT4 = QT_VERSION[0] == '4'
 QT5 = QT_VERSION[0] == '5'

labelme/__init__.py

@@ -118,6 +118,11 @@ def main():
         help='epsilon to find nearest vertex on canvas',
         default=argparse.SUPPRESS,
     )
+    parser.add_argument(
+        '--global-translation-track-size',
+        help='Size in pixel for downsampled frame when estimating global camera translation and rotation. Set to zero to disable the feature. Default: 100',
+        default=argparse.SUPPRESS,
+    )
     parser.add_argument(
         '--ssdmodel',
         help='Filename for SSD multibox weights',
@@ -128,6 +133,28 @@ def main():
         help='Minimum resolution at which to auto annotate. Smaller values result in more memory usage and longer runtime but detect smaller objects. Default: 300',
         default=argparse.SUPPRESS,
     )
+    parser.add_argument(
+        '--ssd-threshold-autoannotation',
+        help='Confidence threshold for auto annotation. Default: 0.15',
+        default=argparse.SUPPRESS,
+    )
+    parser.add_argument(
+        '--ssd-track-crop-factor',
+        help='Region size around the tracked annotation in which to search for objects with SSD, relative to bounding box. Default: 3.0',
+        default=argparse.SUPPRESS,
+    )
+    parser.add_argument(
+        '--ssd-re3-correction-alpha',
+        dest='ssd_re3_correction_alpha',
+        help='Weighting factor for SSD detection coordinates when drift correcting Re³. Default: 0.8',
+        default=argparse.SUPPRESS,
+    )
+    parser.add_argument(
+        '--ssd-re3-min-iou',
+        dest='ssd_re3_correction_min_iou',
+        help='Minimum Jaccard-overlap (IOU) between tracked bounding box and highest confidence SSD detection to consider for Re³ drift correction. Default: 0.8',
+        default=argparse.SUPPRESS,
+    )
     parser.add_argument(
         '--re3model',
         help='Filename for RE3 multibox weights',
@@ -143,6 +170,20 @@ def main():
         help='Device to run Deep Neural Networks on - cpu or cuda',
         default=argparse.SUPPRESS,
     )
+    parser.add_argument(
+        '--autorun',
+        dest='auto_run',
+        action='store_true',
+        help='Run in auto mode with tracking enabled, then exit',
+        default=argparse.SUPPRESS,
+    )
+    parser.add_argument(
+        '--freezekeyframes',
+        dest='freeze_key_frames',
+        action='store_true',
+        help='When auto-tracking, do not change already annotated frames, instead re-init tracker dictionary',
+        default=argparse.SUPPRESS,
+    )
     args = parser.parse_args()
 
     if args.version:

labelme/__main__.py

@@ -29,6 +29,7 @@
 from labelme.widgets import ToolBar
 from labelme.widgets import UniqueLabelQListWidget
 from labelme.widgets import ZoomWidget
+from labelme.widgets import StepsizeWidget
 
 from labelme.trackerre3 import Tracker, trackerAutoAnnotate, trackerInit, trackerDetectFlags, trackerFindGlobalOffset
 
@@ -82,6 +83,7 @@ def __init__(
         # Whether we need to save or not.
         self.dirty = False
         self.image_skip = int(self._config['image_skip'])
+        self.autoMode = False
 
         self._noSelectionSlot = False
 
@@ -164,6 +166,7 @@ def __init__(
         self.file_dock.setWidget(fileListWidget)
 
         self.zoomWidget = ZoomWidget()
+        self.stepsizeWidget = StepsizeWidget(value=self.image_skip)
 
         self.canvas = self.labelList.canvas = Canvas(
             epsilon=self._config['epsilon'],
@@ -381,6 +384,12 @@ def __init__(
                         self.tr('Automatically annotate all object instances visible'),
                         enabled=False)
 
+        startStopAutoMode = action(self.tr('StartStopAutoMode'), self.startStopAutoMode,
+                        shortcuts['startStopAutoMode'], 'icon',
+                        self.tr('Automatically annotate the whole video'),
+                        enabled=False,
+                        checkable=True)
+
         undo = action(self.tr('Undo'), self.undoShapeEdit,
                       shortcuts['undo'], 'undo',
                       self.tr('Undo last add and edit of shape'),
@@ -415,6 +424,15 @@ def __init__(
         )
         self.zoomWidget.setEnabled(False)
 
+        stepsize = QtWidgets.QWidgetAction(self)
+        stepsize.setDefaultWidget(self.stepsizeWidget)
+        self.stepsizeWidget.setWhatsThis(
+            self.tr(
+                'Skip size for iteration through Image.'
+            )
+        )
+        self.stepsizeWidget.setEnabled(True)
+
         zoomIn = action(self.tr('Zoom &In'),
                         functools.partial(self.addZoom, 1.1),
                         shortcuts['zoom_in'], 'zoom-in',
@@ -504,6 +522,8 @@ def __init__(
             createLineStripMode=createLineStripMode,
             zoom=zoom, zoomIn=zoomIn, zoomOut=zoomOut, zoomOrg=zoomOrg,
             fitWindow=fitWindow, fitWidth=fitWidth,
+            startStopAutoMode=startStopAutoMode,
+            stepsize=stepsize,
             zoomActions=zoomActions,
             openNextImg=openNextImg, openPrevImg=openPrevImg, openseg=openseg,
             increase_blend=increase_blend, decrease_blend=decrease_blend,
@@ -553,6 +573,7 @@ def __init__(
                 createLineStripMode,
                 editMode,
                 autoAnnotate,
+                startStopAutoMode,
             ),
             onShapesPresent=(saveAs, hideAll, showAll),
         )
@@ -652,6 +673,8 @@ def __init__(
             zoomOut,
             fitWindow,
             fitWidth,
+            startStopAutoMode,
+            stepsize,
         )
 
         self.statusBar().showMessage(self.tr('%s started.') % __appname__)
@@ -725,12 +748,16 @@ def __init__(
 
         # Callbacks:
         self.zoomWidget.valueChanged.connect(self.paintCanvas)
+        self.stepsizeWidget.valueChanged.connect(self.stepsizeChanged)
 
         self.populateModeActions()
 
         # self.firstStart = True
         # if self.firstStart:
         #    QWhatsThis.enterWhatsThisMode()
+        if self._config['auto_run']:
+            logger.info('Auto Run Engaged')
+            self.queueEvent(functools.partial(self.autoRun))
 
     def menu(self, title, actions=None):
         menu = self.menuBar().addMenu(title)
@@ -799,6 +826,7 @@ def setClean(self):
         self.actions.createPointMode.setEnabled(True)
         self.actions.createLineStripMode.setEnabled(True)
         self.actions.autoAnnotate.setEnabled(True)
+        self.actions.startStopAutoMode.setEnabled(True)
         title = __appname__
         if self.filename is not None:
             title = '{} - {}'.format(title, self.filename)
@@ -1508,6 +1536,17 @@ def loadFile(self, filename=None, reset=True):
                 flags.update(self.labelFile.flags)
         self.loadFlags(flags)
 
+        if self.autoMode and self.tracker_dict and self._config['freeze_key_frames'] and len(self.canvas.shapes)>0:
+            #this frame has annotated labels - we treat it as a key frame and do not change it
+            self.stop_tracker()
+        if self.autoMode and self._config['freeze_key_frames'] and len(self.canvas.shapes)>0:
+            logger.info('Auto Run encountered keyframe %s'%self.filename)
+            selected_shapes=[]
+            for shape in self.canvas.shapes:
+                selected_shapes.append(shape)
+            self.canvas.selectShapes(selected_shapes)
+            self.start_tracker()
+
         if self.tracker_dict:
             otrack_shapes=[]
             for shape in prev_shapes:
@@ -2098,3 +2137,44 @@ def autoAnnotate(self):
         self.loadShapes(track_shapes, replace=False)
         self.setDirty()
 
+    def startStopAutoMode(self):
+        if self.autoMode:
+            self.actions.startStopAutoMode.setChecked(False)
+            self.autoMode = False
+            logger.info('Auto Run Stopped')
+        else:
+            if not self.mayContinue():
+                return
+            self.actions.startStopAutoMode.setChecked(True)
+            self.autoMode = True
+            logger.info('Auto Run Started')
+            self.queueEvent(functools.partial(self.autoStep))
+
+    def stepsizeChanged(self):
+        self.image_skip=int(self.stepsizeWidget.value())
+
+    def autoStep(self):
+        if self.autoMode:
+            if len(self.imageList) <= 0 or self.filename == self.imageList[-1]:
+                self.autoMode = False
+                self.actions.startStopAutoMode.setChecked(False)
+                if self._config['auto_run']:
+                      self.queueEvent(functools.partial(self.autoRunFin))
+
+            else:
+                logger.info('Auto Run: %s'%self.filename)
+                self._openNextImg(skip=self.image_skip)
+                self.queueEvent(functools.partial(self.autoStep))
+
+    def autoRun(self):
+        selected_shapes=[]
+        for shape in self.canvas.shapes:
+            selected_shapes.append(shape)
+        if len(selected_shapes):
+           self.canvas.selectShapes(selected_shapes)
+           self.actions.toggle_tracker.toggle()
+           self.start_tracker()
+        self.startStopAutoMode()
+
+    def autoRunFin(self):
+        self.close()

labelme/app.py

@@ -14,7 +14,7 @@ def main():
     parser.add_argument('dest_annotation_folder')
     parser.add_argument('--source_framerate','-sf',required=True,type=float,help="Framerate of the source annotations")
     parser.add_argument('--dest_framerate','-df',default=30000./1001,type=float,help="Resample to this framerate")
-    parser.add_argument('--dest_template',default='frame_%05d',help="Frame file name template for resulting annotations")
+    parser.add_argument('--dest_template',default='frame_%06d',help="Frame file name template for resulting annotations")
     parser.add_argument('--dest_image_type',default='.jpg',help="Image file name extension written in resulting annotation")
     args = parser.parse_args()
     if not os.path.isdir(args.source_annotation_folder):

labelme/cli/convert_fps.py

@@ -27,13 +27,12 @@ def main():
 
     if data['imageData']:
         imageData = data['imageData']
+        img = utils.img_b64_to_arr(imageData)
     else:
         imagePath = os.path.join(os.path.dirname(json_file), data['imagePath'])
         with open(imagePath, 'rb') as f:
             imageData = f.read()
-            imageData = base64.b64encode(imageData).decode('utf-8')
-    img = utils.img_b64_to_arr(imageData)
-
+            img = utils.img_data_to_arr(imageData)
     label_name_to_value = {'': 0}
     label_texts={0:''}
     for shape in sorted(data['shapes'], key=lambda x: x['label']):

labelme/cli/draw_overlay.py

@@ -1,3 +1,4 @@
+auto_run: false
 auto_save: true
 canvas:
   double_click: null
@@ -21,7 +22,9 @@ flag_dock:
   show: true
 flagmodel: null
 flags: null
-image_skip: 15
+freeze_key_frames: false
+global_translation_track_size: 100
+image_skip: 1
 keep_prev: false
 keep_prev_scale: true
 label_colors: null
@@ -79,6 +82,7 @@ shortcuts:
   save: Ctrl+S
   save_as: Ctrl+Shift+S
   save_to: null
+  startStopAutoMode: Ctrl+Space
   toggle_keep_prev_mode: null
   toggle_tracker:
   - Ctrl+T
@@ -94,7 +98,11 @@ shortcuts:
 show_label_text_field: true
 sort_labels: true
 ssd_min_resolution: 300
+ssd_re3_correction_alpha: 0.8
+ssd_re3_correction_min_iou: 0.8
+ssd_threshold_autoannotation: 0.15
+ssd_track_crop_factor: 3.0
 ssdmodel: null
 store_data: false
 validate_label: null
-version: 5.2.6
+version: 5.2.10

labelme/config/default_config.yaml

@@ -250,7 +250,7 @@ def transform(self, T):
         tpoints=[]
         for p in self.points:
             tp = np.dot(T , [p.x(),p.y(), 1])
-            qp = QtCore.QPoint(tp[0],tp[1])
+            qp = QtCore.QPointF(tp[0],tp[1])
             tpoints.append(qp)
 
         self.points = tpoints
@@ -302,4 +302,4 @@ def __getitem__(self, key):
         return self.points[key]
 
     def __setitem__(self, key, value):
-        self.points[key] = value
+        self.points[key] = value