Generic media support (#539)

- Added `DatasetItem.media` to replace dedicated members for each media type - Added the `PointCloud` media type - Added the `media_type()` method to `Extractor`s - Added merging for all media types, mixed media types for an item or in the dataset produce an error - Datasets can't have mixed media types in items. If such situation occurs, an error is raised (checked during dataset caching/iteration) - Datasets can't change media type using transforms - Extractors must report their media type with the `media_type()` method - Added a new mandatory `media_type` argument to `Dataset.from_iterable`. It has a default value of `Image` for the transition period (to be tracked in #675). - Deprecated `DatasetItem.image`, `.related_images`, `.point_cloud`, `save-images` and `require_images` - Added deprecation messages about annotation classes in `components.extractor` - Suppressed Datumaro deprecation messages when using Datumaro from CLI Co-authored-by: yasakova-anastasia <[email protected]>
openvinotoolkit · Mar 9, 2022 · b7d83d5 · b7d83d5
1 parent b30bf48
commit b7d83d5
Show file tree

Hide file tree

Showing 157 changed files with 3,362 additions and 2,267 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -8,13 +8,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## \[Unreleased\]
 ### Added
-- TBD
+- Support for custom media types, new `PointCloud` media type,
+  `DatasetItem.media` and `.media_as(type)` members
+  (<https://github.com/openvinotoolkit/datumaro/pull/539>)
+- \[API\] A way to request dataset and extractor media type with `media_type`
+  (<https://github.com/openvinotoolkit/datumaro/pull/539>)
 
 ### Changed
 - TBD
 
 ### Deprecated
-- TBD
+- `--save-images` is replaced with `--save-media` in CLI and converter API
+  (<https://github.com/openvinotoolkit/datumaro/pull/539>)
+- \[API\] `image`, `point_cloud` and `related_images` of `DatasetItem` are
+  replaced with `media` and `media_as(type)` members and c-tor parameters
+  (<https://github.com/openvinotoolkit/datumaro/pull/539>)
 
 ### Removed
 - TBD

diff --git a/datumaro/cli/__main__.py b/datumaro/cli/__main__.py
@@ -6,6 +6,7 @@
 import logging as log
 import os.path as osp
 import sys
+import warnings
 
 from ..util.telemetry_utils import (
     close_telemetry_session, init_telemetry_session,
@@ -39,6 +40,10 @@ def init_logger(cls, args=None):
         log.basicConfig(format='%(asctime)s %(levelname)s: %(message)s',
             level=args.loglevel)
 
+        # Suppress own deprecation warnings
+        warnings.filterwarnings('ignore', category=DeprecationWarning,
+            module=r'datumaro\..*')
+
     @staticmethod
     def _define_loglevel_option(parser):
         parser.add_argument('--loglevel', type=loglevel, default='info',

diff --git a/datumaro/cli/commands/download.py b/datumaro/cli/commands/download.py
@@ -47,7 +47,7 @@ def build_parser(parser_ctor=argparse.ArgumentParser):
         |n
         Examples:|n
         - Download the MNIST dataset:|n
-        |s|s%(prog)s -i tfds:mnist -- --save-images|n
+        |s|s%(prog)s -i tfds:mnist -- --save-media|n
         |n
         - Download the VOC 2012 dataset, saving only the annotations in the COCO
           format into a specific directory:|n

diff --git a/datumaro/cli/contexts/project/__init__.py b/datumaro/cli/contexts/project/__init__.py
@@ -371,10 +371,10 @@ def filter_command(args):
 
                 # Source might be missing in the working dir, so we specify
                 # the output directory.
-                # We specify save_images here as a heuristic. It can probably
+                # We specify save_media here as a heuristic. It can probably
                 # be improved by checking if there are images in the dataset
                 # directory.
-                dataset.save(project.source_data_dir(target), save_images=True)
+                dataset.save(project.source_data_dir(target), save_media=True)
 
             log.info("Finished")
         else:
@@ -389,7 +389,7 @@ def filter_command(args):
             dst_dir = osp.abspath(dst_dir)
 
             dataset.filter(filter_expr, *filter_args)
-            dataset.save(dst_dir, save_images=True)
+            dataset.save(dst_dir, save_media=True)
 
             log.info("Results have been saved to '%s'" % dst_dir)
 
@@ -557,10 +557,10 @@ def transform_command(args):
 
                 # Source might be missing in the working dir, so we specify
                 # the output directory
-                # We specify save_images here as a heuristic. It can probably
+                # We specify save_media here as a heuristic. It can probably
                 # be improved by checking if there are images in the dataset
                 # directory.
-                dataset.save(project.source_data_dir(target), save_images=True)
+                dataset.save(project.source_data_dir(target), save_media=True)
 
             log.info("Finished")
         else:
@@ -575,7 +575,7 @@ def transform_command(args):
             dst_dir = osp.abspath(dst_dir)
 
             dataset.transform(args.transform, **extra_args)
-            dataset.save(dst_dir, save_images=True)
+            dataset.save(dst_dir, save_media=True)
 
             log.info("Results have been saved to '%s'" % dst_dir)
 

diff --git a/datumaro/cli/contexts/project/diff.py b/datumaro/cli/contexts/project/diff.py
@@ -14,6 +14,8 @@
 import cv2
 import numpy as np
 
+from datumaro.components.media import Image
+
 with warnings.catch_warnings():
     warnings.simplefilter("ignore")
     import tensorboardX as tb
@@ -121,7 +123,10 @@ def save(self, a: IDataset, b: IDataset):
             self.update_mask_confusion(mask_diff)
 
             self.save_item_label_diff(item_a, item_b, label_diff)
-            self.save_item_bbox_diff(item_a, item_b, bbox_diff)
+
+            if a.media_type() and issubclass(a.media_type(), Image) and \
+                    b.media_type() and issubclass(b.media_type(), Image):
+                self.save_item_bbox_diff(item_a, item_b, bbox_diff)
 
         if len(self.label_confusion_matrix) != 0:
             self.save_conf_matrix(self.label_confusion_matrix,
@@ -243,11 +248,11 @@ def save_item_bbox_diff(self, item_a, item_b, diff):
         _, mispred, a_unmatched, b_unmatched = diff
 
         if 0 < len(a_unmatched) + len(b_unmatched) + len(mispred):
-            if not item_a.has_image or not item_a.image.has_data:
+            if not isinstance(item_a.media, Image) or not item_a.media.has_data:
                 log.warning("Item %s: item has no image data, "
                     "it will be skipped" % (item_a.id))
                 return
-            img_a = item_a.image.data.copy()
+            img_a = item_a.media.data.copy()
             img_b = img_a.copy()
             for a_bbox, b_bbox in mispred:
                 self.draw_bbox(img_a, a_bbox, self.get_a_label(a_bbox.label),

diff --git a/datumaro/components/converter.py b/datumaro/components/converter.py
@@ -8,16 +8,17 @@
 import os
 import os.path as osp
 import shutil
+import warnings
 
 from attrs import define, field
 import attr
 
 from datumaro.components.cli_plugin import CliPlugin
 from datumaro.components.errors import (
-    AnnotationExportError, DatumaroError, ItemExportError,
+    AnnotationExportError, DatasetExportError, DatumaroError, ItemExportError,
 )
 from datumaro.components.extractor import DatasetItem, IExtractor
-from datumaro.components.media import Image
+from datumaro.components.media import Image, PointCloud
 from datumaro.components.progress_reporting import (
     NullProgressReporter, ProgressReporter,
 )
@@ -89,8 +90,16 @@ class Converter(CliPlugin):
     @classmethod
     def build_cmdline_parser(cls, **kwargs):
         parser = super().build_cmdline_parser(**kwargs)
+
+        # Deprecated
         parser.add_argument('--save-images', action='store_true',
-            help="Save images (default: %(default)s)")
+            default=None,
+            help="(Deprecated. Use --save-media instead) "
+                "Save images (default: False)")
+
+        parser.add_argument('--save-media', action='store_true',
+            default=None, # TODO: remove default once save-images is removed
+            help="Save media (default: False)")
         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 ''))
@@ -138,7 +147,8 @@ def apply(self):
         raise NotImplementedError("Should be implemented in a subclass")
 
     def __init__(self, extractor: IExtractor, save_dir: str, *,
-            save_images: bool = False,
+            save_images=None, # Deprecated
+            save_media: Optional[bool] = None,
             image_ext: Optional[str] = None,
             default_image_ext: Optional[str] = None,
             save_dataset_meta: bool = False,
@@ -147,7 +157,20 @@ def __init__(self, extractor: IExtractor, save_dir: str, *,
         assert default_image_ext
         self._default_image_ext = default_image_ext
 
-        self._save_images = save_images
+        if save_images is not None and save_media is not None:
+            raise DatasetExportError("Can't use both 'save-media' and "
+                "'save-images'")
+
+        if save_media is not None:
+            self._save_media = save_media
+        elif save_images is not None:
+            self._save_media = save_images
+            warnings.warn("'save-images' is deprecated and will be "
+                "removed in future. Use 'save-media' instead.",
+                DeprecationWarning, stacklevel=2)
+        else:
+            self._save_media = False
+
         self._image_ext = image_ext
 
         self._extractor = extractor
@@ -168,8 +191,8 @@ def __init__(self, extractor: IExtractor, save_dir: str, *,
     def _find_image_ext(self, item: Union[DatasetItem, Image]):
         src_ext = None
 
-        if isinstance(item, DatasetItem) and item.has_image:
-            src_ext = item.image.ext
+        if isinstance(item, DatasetItem) and isinstance(item.media, Image):
+            src_ext = item.media.ext
         elif isinstance(item, Image):
             src_ext = item.ext
 
@@ -192,7 +215,7 @@ def _save_image(self, item, path=None, *,
         assert not ((subdir or name or basedir) and path), \
             "Can't use both subdir or name or basedir and path arguments"
 
-        if not item.has_image or not item.image.has_data:
+        if not isinstance(item.media, Image) or not item.media.has_data:
             log.warning("Item '%s' has no image", item.id)
             return
 
@@ -201,14 +224,14 @@ def _save_image(self, item, path=None, *,
             self._make_image_filename(item, name=name, subdir=subdir))
         path = osp.abspath(path)
 
-        item.image.save(path)
+        item.media.save(path)
 
     def _save_point_cloud(self, item=None, path=None, *,
             name=None, subdir=None, basedir=None):
         assert not ((subdir or name or basedir) and path), \
             "Can't use both subdir or name or basedir and path arguments"
 
-        if not item.point_cloud:
+        if not item.media or not isinstance(item.media, PointCloud):
             log.warning("Item '%s' has no pcd", item.id)
             return
 
@@ -218,9 +241,9 @@ def _save_point_cloud(self, item=None, path=None, *,
         path = osp.abspath(path)
 
         os.makedirs(osp.dirname(path), exist_ok=True)
-        if item.point_cloud and osp.isfile(item.point_cloud):
-            if item.point_cloud != path:
-                shutil.copyfile(item.point_cloud, path)
+        if item.media and osp.isfile(item.media.path):
+            if item.media.path != path:
+                shutil.copyfile(item.media.path, path)
 
     def _save_meta_file(self, path):
         save_meta_file(path, self._extractor.categories())