docstrings

Author: Matt Sokoloff

examples/integrations/detectron2/coco_utils.py

@@ -47,7 +47,7 @@ def partition_coco(coco_instance_data, coco_panoptic_data = None, splits = None)
     return partitions
 
 
-def visualize_coco_examples(coco_examples, metadata_catalog, scale = 1.0, max_images = 5, resize_dims = (512,768)):
+def visualize_coco_examples(coco_examples, metadata_catalog, scale = 1.0, max_images = 5, resize_dims = (768, 512)):
     images = []
     for idx, example in enumerate(coco_examples):
         if idx > max_images:

labelbox/data/serialization/coco/converter.py

@@ -1,8 +1,3 @@
-# Mapping can be used to select classes as instsance or segmentation
-# By default polygon will be instance and segmentation will be semantic
-
-# Will use subclasses too..
-# checklist and radio will be turned into underscore delimitated names
 from typing import Dict, Any
 import os
 
@@ -14,27 +9,37 @@
 def create_path_if_not_exists(path: str):
     if not os.path.exists(path):
         os.makedirs(path)
+    elif os.path.listdir(path):
+        raise ValueError(f"Directory `{path}`` must be empty.")
 
 
 def validate_path(path, name):
     if not os.path.exists(path):
-        raise ValueError(f"{name} {path} must exist")
+        raise ValueError(f"{name} `{path}` must exist")
 
 
 class COCOConverter:
     """
-    Note that this class is only compatible with image data.. it will ignore other data types.
-    # TODO: Filter out video annotations..
-
+    Class for convertering between coco and labelbox formats
+    Note that this class is only compatible with image data.
 
-    # Note that these functions are not compatible with subclasses currently.
-    # If you have subclasses you will have to manually flatten them before using the converter.
+    Subclasses are currently ignored.
+    To use subclasses, manually flatten them before using the converter.
     """
 
     def serialize_instances(labels: LabelCollection,
                             image_root: str) -> Dict[str, Any]:
         """
-        Compatible with masks, polygons, and masks. Will turn masks into individual instances
+        Convert a Labelbox LabelCollection into an mscoco dataset.
+        This function will only convert masks, polygons, and rectangles.
+        Masks will be converted into individual instances.
+        Use deserialize_panoptic to prevent masks from being split apart.
+
+        Args:
+            labels: A collection of labels to convert
+            image_root: Where to save images to
+        Returns:
+            A dictionary containing labels in the coco object format.
         """
         create_path_if_not_exists(image_root)
         return CocoInstanceDataset.from_common(labels=labels,
@@ -43,8 +48,19 @@ def serialize_instances(labels: LabelCollection,
     def serialize_panoptic(labels: LabelCollection, image_root: str,
                            mask_root: str, all_stuff: bool = False) -> Dict[str, Any]:
         """
-        All stuff turns every class into segmentation classes
+        Convert a Labelbox LabelCollection into an mscoco dataset.
+        This function will only convert masks, polygons, and rectangles.
+        Masks will be converted into individual instances.
+        Use deserialize_panoptic to prevent masks from being split apart.
 
+        Args:
+            labels: A collection of labels to convert
+            image_root: Where to save images to
+            mask_root: Where to save segmentation masks to
+            all_stuff: If rectangle or polygon annotations are encountered, they will be treated as instances.
+                To convert them to stuff class set `all_stuff=True`.
+        Returns:
+            A dictionary containing labels in the coco panoptic format.
         """
         create_path_if_not_exists(image_root)
         create_path_if_not_exists(mask_root)
@@ -55,6 +71,16 @@ def serialize_panoptic(labels: LabelCollection, image_root: str,
     @classmethod
     def deserialize_panoptic(cls, json_data: Dict[str, Any], image_root: str,
                              mask_root: str) -> LabelGenerator:
+        """
+        Convert coco panoptic data into the labelbox format (as a LabelGenerator).
+
+        Args:
+            json_data: panoptic data as a dict
+            image_root: Path to local images that are referenced by the panoptic json
+            mask_root: Path to local segmentation masks that are referenced by the panoptic json
+        Returns:
+            LabelGenerator
+        """
         validate_path(image_root, 'image_root')
         validate_path(mask_root, 'mask_root')
         objs = CocoPanopticDataset(**json_data)
@@ -64,6 +90,15 @@ def deserialize_panoptic(cls, json_data: Dict[str, Any], image_root: str,
     @classmethod
     def deserialize_instances(cls, json_data: Dict[str, Any],
                               image_root: str) -> LabelGenerator:
+        """
+        Convert coco object data into the labelbox format (as a LabelGenerator).
+
+        Args:
+            json_data: coco object data as a dict
+            image_root: Path to local images that are referenced by the coco object json
+        Returns:
+            LabelGenerator
+        """
         validate_path(image_root, 'image_root')
         objs = CocoInstanceDataset(**json_data)
         gen = objs.to_common(image_root)

labelbox/data/serialization/coco/instance_dataset.py

@@ -82,8 +82,8 @@ def segmentations_to_common(class_annotations, class_name):
     return annotations
 
 
-def process_label(label: Label, idx, image_root):
-    annot_idx = idx * 10000
+def process_label(label: Label, idx, image_root, max_annotations_per_image = 10000):
+    annot_idx = idx * max_annotations_per_image
     image_id = get_image_id(label, idx)
     image = get_image(label, image_root, image_id)
     coco_annotations = []

labelbox/data/serialization/coco/panoptic_dataset.py

@@ -30,7 +30,7 @@ def vector_to_coco_segment_info(canvas: np.ndarray,
     return SegmentInfo(id=annotation_idx,
                        category_id=category_id,
                        area=shapely.area,
-                       bbox=[xmin, ymin, xmax - xmin, ymax - ymin])
+                       bbox=[xmin, ymin, xmax - xmin, ymax - ymin]), canvas
 
 
 def mask_to_coco_segment_info(canvas: np.ndarray, annotation, annotation_idx: int, category_id):
@@ -69,13 +69,13 @@ def process_label(label: Label, idx: Union[int, str], image_root, mask_root, all
                 is_thing[annotation.name] = 0
 
             elif isinstance(annotation.value, (Polygon, Rectangle)):
-                segments.append(
-                    vector_to_coco_segment_info(
+                segment, canvas = vector_to_coco_segment_info(
                         canvas,
                         annotation,
                         annotation_idx=(class_idx if all_stuff else annotation_idx) + 1,
                         image=image,
-                        category_id=categories[annotation.name]))
+                        category_id=categories[annotation.name])
+                segments.append(segment)
                 is_thing[annotation.name] = 1 - int(all_stuff)
 
     mask_file = image.file_name.replace('.jpg', '.png')