[Datumaro] Mandatory sorting for black color in voc masks

CHANGELOG.md

@@ -28,6 +28,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 - Issue loading openvino models for semi-automatic and automatic annotation (<https://github.com/opencv/cvat/pull/1996>)
 - Basic functions of CVAT works without activated nuclio dashboard
+- Fixed a case in which exported masks could have wrong color order (<https://github.com/opencv/cvat/issues/2032>)
 - Fixed error with creating task with labels with the same name (<https://github.com/opencv/cvat/pull/2031>)
 - Django RQ dashboard view (<https://github.com/opencv/cvat/pull/2069>)
 

datumaro/datumaro/plugins/voc_format/converter.py

@@ -47,7 +47,7 @@ def _write_xml_bbox(bbox, parent_elem):
     return bbox_elem
 
 
-LabelmapType = Enum('LabelmapType', ['voc', 'source', 'guess'])
+LabelmapType = Enum('LabelmapType', ['voc', 'source'])
 
 class VocConverter(Converter):
     DEFAULT_IMAGE_EXT = VocPath.IMAGE_EXT
@@ -102,6 +102,8 @@ def __init__(self, extractor, save_dir,
         self._apply_colormap = apply_colormap
         self._allow_attributes = allow_attributes
 
+        if label_map is None:
+            label_map = LabelmapType.source
         self._load_categories(label_map)
 
     def apply(self):
@@ -446,7 +448,7 @@ def save_label_map(self):
         path = osp.join(self._save_dir, VocPath.LABELMAP_FILE)
         write_label_map(path, self._label_map)
 
-    def _load_categories(self, label_map_source=None):
+    def _load_categories(self, label_map_source):
         if label_map_source == LabelmapType.voc.name:
             # use the default VOC colormap
             label_map = make_voc_label_map()
@@ -456,71 +458,54 @@ def _load_categories(self, label_map_source=None):
             # generate colormap for input labels
             labels = self._extractor.categories() \
                 .get(AnnotationType.label, LabelCategories())
-            label_map = OrderedDict()
-            label_map['background'] = [None, [], []]
-            for item in labels.items:
-                label_map[item.name] = [None, [], []]
+            label_map = OrderedDict((item.name, [None, [], []])
+                for item in labels.items)
 
         elif label_map_source == LabelmapType.source.name and \
                 AnnotationType.mask in self._extractor.categories():
             # use source colormap
             labels = self._extractor.categories()[AnnotationType.label]
             colors = self._extractor.categories()[AnnotationType.mask]
             label_map = OrderedDict()
-            has_black = False
             for idx, item in enumerate(labels.items):
                 color = colors.colormap.get(idx)
-                if idx is not None:
-                    if color == (0, 0, 0):
-                        has_black = True
+                if color is not None:
                     label_map[item.name] = [color, [], []]
-            if not has_black and 'background' not in label_map:
-                label_map['background'] = [(0, 0, 0), [], []]
-                label_map.move_to_end('background', last=False)
-
-        elif label_map_source in [LabelmapType.guess.name, None]:
-            # generate colormap for union of VOC and input dataset labels
-            label_map = make_voc_label_map()
-
-            rebuild_colormap = False
-            source_labels = self._extractor.categories() \
-                .get(AnnotationType.label, LabelCategories())
-            for label in source_labels.items:
-                if label.name not in label_map:
-                    rebuild_colormap = True
-                if label.attributes or label.name not in label_map:
-                    label_map[label.name] = [None, [], label.attributes]
-
-            if rebuild_colormap:
-                for item in label_map.values():
-                    item[0] = None
 
         elif isinstance(label_map_source, dict):
-            label_map = label_map_source
+            label_map = OrderedDict(
+                sorted(label_map_source.items(), key=lambda e: e[0]))
 
         elif isinstance(label_map_source, str) and osp.isfile(label_map_source):
             label_map = parse_label_map(label_map_source)
 
-            has_black = find(label_map.items(),
-                lambda e: e[0] == 'background' or e[1][0] == (0, 0, 0))
-            if not has_black and 'background' not in label_map:
-                label_map['background'] = [(0, 0, 0), [], []]
-                label_map.move_to_end('background', last=False)
-
         else:
             raise Exception("Wrong labelmap specified, "
                 "expected one of %s or a file path" % \
                 ', '.join(t.name for t in LabelmapType))
 
+        # There must always be a label with color (0, 0, 0) at index 0
+        bg_label = find(label_map.items(), lambda x: x[1][0] == (0, 0, 0))
+        if bg_label is not None:
+            bg_label = bg_label[0]
+        else:
+            bg_label = 'background'
+            if bg_label not in label_map:
+                has_colors = any(v[0] is not None for v in label_map.values())
+                color = (0, 0, 0) if has_colors else None
+                label_map[bg_label] = [color, [], []]
+        label_map.move_to_end(bg_label, last=False)
+
         self._categories = make_voc_categories(label_map)
 
-        self._label_map = label_map
+        # Update colors with assigned values
         colormap = self._categories[AnnotationType.mask].colormap
         for label_id, color in colormap.items():
             label_desc = label_map[
                 self._categories[AnnotationType.label].items[label_id].name]
             label_desc[0] = color
 
+        self._label_map = label_map
         self._label_id_mapping = self._make_label_id_map()
 
     def _is_label(self, s):

datumaro/datumaro/plugins/voc_format/format.py

@@ -137,6 +137,9 @@ def parse_label_map(path):
             label_desc = line.strip().split(':')
             name = label_desc[0]
 
+            if name in label_map:
+                raise ValueError("Label '%s' is already defined" % name)
+
             if 1 < len(label_desc) and len(label_desc[1]) != 0:
                 color = label_desc[1].split(',')
                 assert len(color) == 3, \
@@ -173,7 +176,6 @@ def write_label_map(path, label_map):
 
             f.write('%s\n' % ':'.join([label_name, color_rgb, parts, actions]))
 
-# pylint: disable=pointless-statement
 def make_voc_categories(label_map=None):
     if label_map is None:
         label_map = make_voc_label_map()
@@ -190,16 +192,15 @@ def make_voc_categories(label_map=None):
         label_categories.add(part)
     categories[AnnotationType.label] = label_categories
 
-    has_colors = sum(v[0] is not None for v in label_map.values())
-    if not has_colors:
+    has_colors = any(v[0] is not None for v in label_map.values())
+    if not has_colors: # generate new colors
         colormap = generate_colormap(len(label_map))
-    else:
+    else: # only copy defined colors
         label_id = lambda label: label_categories.find(label)[0]
         colormap = { label_id(name): desc[0]
-            for name, desc in label_map.items() }
+            for name, desc in label_map.items() if desc[0] is not None }
     mask_categories = MaskCategories(colormap)
-    mask_categories.inverse_colormap # force init
+    mask_categories.inverse_colormap # pylint: disable=pointless-statement
     categories[AnnotationType.mask] = mask_categories
 
     return categories
-# pylint: enable=pointless-statement

datumaro/tests/test_voc_format.py

@@ -472,53 +472,6 @@ def categories(self):
                 partial(VocConverter.convert, label_map='voc'),
                 test_dir, target_dataset=DstExtractor())
 
-    def test_dataset_with_guessed_labelmap(self):
-        class SrcExtractor(TestExtractorBase):
-            def __iter__(self):
-                yield DatasetItem(id=1, annotations=[
-                    Bbox(2, 3, 4, 5, label=0, id=1),
-                    Bbox(1, 2, 3, 4, label=1, id=2),
-                ])
-
-            def categories(self):
-                label_cat = LabelCategories()
-                label_cat.add(VOC.VocLabel(1).name)
-                label_cat.add('non_voc_label')
-                return {
-                    AnnotationType.label: label_cat,
-                }
-
-        class DstExtractor(TestExtractorBase):
-            def __iter__(self):
-                yield DatasetItem(id=1, annotations=[
-                    Bbox(2, 3, 4, 5, label=self._label(VOC.VocLabel(1).name),
-                        id=1, group=1, attributes={
-                            'truncated': False,
-                            'difficult': False,
-                            'occluded': False,
-                        }
-                    ),
-                    Bbox(1, 2, 3, 4, label=self._label('non_voc_label'),
-                        id=2, group=2, attributes={
-                            'truncated': False,
-                            'difficult': False,
-                            'occluded': False,
-                        }
-                    ),
-                ])
-
-            def categories(self):
-                label_map = VOC.make_voc_label_map()
-                label_map['non_voc_label'] = [None, [], []]
-                for label_desc in label_map.values():
-                    label_desc[0] = None # rebuild colormap
-                return VOC.make_voc_categories(label_map)
-
-        with TestDir() as test_dir:
-            self._test_save_and_load(SrcExtractor(),
-                partial(VocConverter.convert, label_map='guess'),
-                test_dir, target_dataset=DstExtractor())
-
     def test_dataset_with_source_labelmap_undefined(self):
         class SrcExtractor(TestExtractorBase):
             def __iter__(self):
@@ -602,8 +555,8 @@ def __iter__(self):
 
             def categories(self):
                 label_map = OrderedDict()
-                label_map['label_1'] = [(1, 2, 3), [], []]
                 label_map['background'] = [(0, 0, 0), [], []]
+                label_map['label_1'] = [(1, 2, 3), [], []]
                 label_map['label_2'] = [(3, 2, 1), [], []]
                 return VOC.make_voc_categories(label_map)
 
@@ -616,11 +569,11 @@ def test_dataset_with_fixed_labelmap(self):
         class SrcExtractor(TestExtractorBase):
             def __iter__(self):
                 yield DatasetItem(id=1, annotations=[
-                    Bbox(2, 3, 4, 5, label=0, id=1),
-                    Bbox(1, 2, 3, 4, label=1, id=2, group=2,
+                    Bbox(2, 3, 4, 5, label=self._label('foreign_label'), id=1),
+                    Bbox(1, 2, 3, 4, label=self._label('label'), id=2, group=2,
                         attributes={'act1': True}),
-                    Bbox(2, 3, 4, 5, label=2, id=3, group=2),
-                    Bbox(2, 3, 4, 6, label=3, id=4, group=2),
+                    Bbox(2, 3, 4, 5, label=self._label('label_part1'), group=2),
+                    Bbox(2, 3, 4, 6, label=self._label('label_part2'), group=2),
                 ])
 
             def categories(self):
@@ -633,14 +586,19 @@ def categories(self):
                     AnnotationType.label: label_cat,
                 }
 
-        label_map = {
-            'label': [None, ['label_part1', 'label_part2'], ['act1', 'act2']]
-        }
+        label_map = OrderedDict([
+            ('label', [None, ['label_part1', 'label_part2'], ['act1', 'act2']])
+        ])
+
+        dst_label_map = OrderedDict([
+            ('background', [None, [], []]),
+            ('label', [None, ['label_part1', 'label_part2'], ['act1', 'act2']])
+        ])
 
         class DstExtractor(TestExtractorBase):
             def __iter__(self):
                 yield DatasetItem(id=1, annotations=[
-                    Bbox(1, 2, 3, 4, label=0, id=1, group=1,
+                    Bbox(1, 2, 3, 4, label=self._label('label'), id=1, group=1,
                         attributes={
                             'act1': True,
                             'act2': False,
@@ -649,12 +607,12 @@ def __iter__(self):
                             'occluded': False,
                         }
                     ),
-                    Bbox(2, 3, 4, 5, label=1, group=1),
-                    Bbox(2, 3, 4, 6, label=2, group=1),
+                    Bbox(2, 3, 4, 5, label=self._label('label_part1'), group=1),
+                    Bbox(2, 3, 4, 6, label=self._label('label_part2'), group=1),
                 ])
 
             def categories(self):
-                return VOC.make_voc_categories(label_map)
+                return VOC.make_voc_categories(dst_label_map)
 
         with TestDir() as test_dir:
             self._test_save_and_load(SrcExtractor(),