[Datumaro] Image control in converters (#1799)

* Update converter interface, add image control

* Update bindings

* update changelog

* Fix build

CHANGELOG.md

@@ -55,6 +55,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Appearance block in attribute annotation mode (<https://github.com/opencv/cvat/pull/1820>)
 - Keyframe navigations and some switchers in attribute annotation mode (<https://github.com/opencv/cvat/pull/1820>)
 - [Datumaro] Added `convert` command to convert datasets directly (<https://github.com/opencv/cvat/pull/1837>)
+- [Datumaro] Added an option to specify image extension when exporting datasets (<https://github.com/opencv/cvat/pull/1799>)
+- [Datumaro] Added image copying when exporting datasets, if possible (<https://github.com/opencv/cvat/pull/1799>)
 
 ### Changed
 - Removed information about e-mail from the basic user information (<https://github.com/opencv/cvat/pull/1627>)
@@ -65,6 +67,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added support for attributes in VOC XML format (https://github.com/opencv/cvat/pull/1792)
 - Added annotation attributes in COCO format (https://github.com/opencv/cvat/pull/1782)
 - Colorized object items in the side panel (<https://github.com/opencv/cvat/pull/1753>)
+- [Datumaro] Annotation-less files are not generated anymore in COCO format, unless tasks explicitly requested (<https://github.com/opencv/cvat/pull/1799>)
 
 ### Deprecated
 -

cvat/apps/dataset_manager/formats/coco.py

@@ -18,9 +18,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('coco_instances',
-            save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('coco_instances').convert(extractor,
+            save_dir=temp_dir, save_images=save_images)
 
         make_zip_archive(temp_dir, dst_file)
 

cvat/apps/dataset_manager/formats/datumaro/__init__.py

@@ -48,11 +48,10 @@ def _save_image_info(save_dir, task_data):
 
     def _export(self, task_data, save_dir, save_images=False):
         dataset = CvatTaskDataExtractor(task_data, include_images=save_images)
-        converter = dm_env.make_converter('datumaro_project',
-            save_images=save_images,
-            config={ 'project_name': task_data.db_task.name, }
+        dm_env.converters.get('datumaro_project').convert(dataset,
+            save_dir=save_dir, save_images=save_images,
+            project_config={ 'project_name': task_data.db_task.name, }
         )
-        converter(dataset, save_dir=save_dir)
 
         project = Project.load(save_dir)
         target_dir = project.config.project_dir

cvat/apps/dataset_manager/formats/labelme.py

@@ -19,8 +19,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('label_me', save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('label_me').convert(extractor, save_dir=temp_dir,
+            save_images=save_images)
 
         make_zip_archive(temp_dir, dst_file)
 

cvat/apps/dataset_manager/formats/mask.py

@@ -26,10 +26,9 @@ def _export(dst_file, task_data, save_images=False):
     extractor = extractor.transform(envt.get('merge_instance_segments'))
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('voc_segmentation',
-            apply_colormap=True, label_map=make_colormap(task_data),
-            save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('voc_segmentation').convert(extractor,
+            save_dir=temp_dir, save_images=save_images,
+            apply_colormap=True, label_map=make_colormap(task_data))
 
         make_zip_archive(temp_dir, dst_file)
 
@@ -68,8 +67,9 @@ def normalize_label(label):
 def make_colormap(task_data):
     labels = sorted([label['name']
         for _, label in task_data.meta['task']['labels']])
-    if 'background' not in labels:
-        labels.insert(0, 'background')
+    if 'background' in labels:
+        labels.remove('background')
+    labels.insert(0, 'background')
 
     predefined = parse_default_colors()
 

cvat/apps/dataset_manager/formats/mot.py

@@ -19,9 +19,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('mot_seq_gt',
-            save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('mot_seq_gt').convert(extractor,
+            save_dir=temp_dir, save_images=save_images)
 
         make_zip_archive(temp_dir, dst_file)
 

cvat/apps/dataset_manager/formats/pascal_voc.py

@@ -24,9 +24,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('voc', label_map='source',
-            save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('voc').convert(extractor,
+            save_dir=temp_dir, save_images=save_images, label_map='source')
 
         make_zip_archive(temp_dir, dst_file)
 

cvat/apps/dataset_manager/formats/tfrecord.py

@@ -27,9 +27,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('tf_detection_api',
-            save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('tf_detection_api').convert(extractor,
+            save_dir=temp_dir, save_images=save_images)
 
         make_zip_archive(temp_dir, dst_file)
 

cvat/apps/dataset_manager/formats/yolo.py

@@ -23,8 +23,8 @@ def _export(dst_file, task_data, save_images=False):
     extractor = CvatTaskDataExtractor(task_data, include_images=save_images)
     extractor = Dataset.from_extractors(extractor) # apply lazy transforms
     with TemporaryDirectory() as temp_dir:
-        converter = dm_env.make_converter('yolo', save_images=save_images)
-        converter(extractor, save_dir=temp_dir)
+        dm_env.converters.get('yolo').convert(extractor,
+            save_dir=temp_dir, save_images=save_images)
 
         make_zip_archive(temp_dir, dst_file)
 

datumaro/datumaro/cli/contexts/project/__init__.py

@@ -340,9 +340,9 @@ def export_command(args):
         raise CliException("Converter for format '%s' is not found" % \
             args.format)
 
-    if hasattr(converter, 'from_cmdline'):
-        extra_args = converter.from_cmdline(args.extra_args)
-        converter = converter(**extra_args)
+    extra_args = converter.from_cmdline(args.extra_args)
+    def converter_proxy(extractor, save_dir):
+        return converter.convert(extractor, save_dir, **extra_args)
 
     filter_args = FilterModes.make_filter_args(args.filter_mode)
 
@@ -352,7 +352,7 @@ def export_command(args):
     log.info("Exporting the project...")
     dataset.export_project(
         save_dir=dst_dir,
-        converter=converter,
+        converter=converter_proxy,
         filter_expr=args.filter,
         **filter_args)
     log.info("Project exported to '%s' as '%s'" % \

datumaro/datumaro/components/converter.py

@@ -3,17 +3,77 @@
 #
 # SPDX-License-Identifier: MIT
 
-class Converter:
-    def __init__(self, cmdline_args=None):
-        pass
+import logging as log
+import os
+import os.path as osp
+import shutil
 
-    def __call__(self, extractor, save_dir):
-        raise NotImplementedError()
+from datumaro.components.cli_plugin import CliPlugin
+from datumaro.util.image import save_image
 
-    def _parse_cmdline(self, cmdline):
-        parser = self.build_cmdline_parser()
 
-        if len(cmdline) != 0 and cmdline[0] == '--':
-            cmdline = cmdline[1:]
-        args = parser.parse_args(cmdline)
-        return vars(args)
+class IConverter:
+    @classmethod
+    def convert(cls, extractor, save_dir, **options):
+        raise NotImplementedError("Should be implemented in a subclass")
+
+class Converter(IConverter, CliPlugin):
+    DEFAULT_IMAGE_EXT = None
+
+    @classmethod
+    def build_cmdline_parser(cls, **kwargs):
+        parser = super().build_cmdline_parser(**kwargs)
+        parser.add_argument('--save-images', action='store_true',
+            help="Save images (default: %(default)s)")
+        parser.add_argument('--image-ext', default=None,
+            help="Image extension (default: keep or use format default%s)" % \
+                (' ' + cls.DEFAULT_IMAGE_EXT if cls.DEFAULT_IMAGE_EXT else ''))
+
+        return parser
+
+    @classmethod
+    def convert(cls, extractor, save_dir, **options):
+        converter = cls(extractor, save_dir, **options)
+        return converter.apply()
+
+    def apply(self):
+        raise NotImplementedError("Should be implemented in a subclass")
+
+    def __init__(self, extractor, save_dir, save_images=False,
+            image_ext=None, default_image_ext=None):
+        default_image_ext = default_image_ext or self.DEFAULT_IMAGE_EXT
+        assert default_image_ext
+        self._default_image_ext = default_image_ext
+
+        self._save_images = save_images
+        self._image_ext = image_ext
+
+        self._extractor = extractor
+        self._save_dir = save_dir
+
+    def _find_image_ext(self, item):
+        src_ext = None
+        if item.has_image:
+            src_ext = osp.splitext(osp.basename(item.image.path))[1]
+
+        return self._image_ext or src_ext or self._default_image_ext
+
+    def _make_image_filename(self, item):
+        return item.id + self._find_image_ext(item)
+
+    def _save_image(self, item, path=None):
+        image = item.image.data
+        if image is None:
+            log.warning("Item '%s' has no image", item.id)
+            return item.image.path
+
+        path = path or self._make_image_filename(item)
+
+        src_ext = osp.splitext(osp.basename(item.image.path))[1]
+        dst_ext = osp.splitext(osp.basename(path))[1]
+
+        os.makedirs(osp.dirname(path), exist_ok=True)
+        if src_ext == dst_ext and osp.isfile(item.image.path):
+            shutil.copyfile(item.image.path, path)
+        else:
+            save_image(path, image)

datumaro/datumaro/components/project.py

@@ -624,9 +624,8 @@ def save(self, save_dir=None, merge=False, recursive=True,
 
             if merge:
                 # merge and save the resulting dataset
-                converter = self.env.make_converter(
-                    DEFAULT_FORMAT, **converter_kwargs)
-                converter(self, dataset_save_dir)
+                self.env.converters.get(DEFAULT_FORMAT).convert(
+                    self, dataset_save_dir, **converter_kwargs)
             else:
                 if recursive:
                     # children items should already be updated
@@ -635,9 +634,8 @@ def save(self, save_dir=None, merge=False, recursive=True,
                         if isinstance(source, ProjectDataset):
                             source.save(**converter_kwargs)
 
-                converter = self.env.make_converter(
-                    DEFAULT_FORMAT, **converter_kwargs)
-                converter(self.iterate_own(), dataset_save_dir)
+                self.env.converters.get(DEFAULT_FORMAT).convert(
+                    self.iterate_own(), dataset_save_dir, **converter_kwargs)
 
             project.save(save_dir)
         except BaseException: