feat: New labels for CVAT annotation (#314)

* New labels and utility for CVAT annotation

Signed-off-by: Christoph Auer <[email protected]>

* Add EMPTY_VALUE to TextItem

Signed-off-by: Christoph Auer <[email protected]>

* Small fixes

Signed-off-by: Christoph Auer <[email protected]>

* align usage of content layer param (#326)

* align usage of content layer param

Signed-off-by: Panos Vagenas <[email protected]>

* parametrize content layers in visualizers

Signed-off-by: Panos Vagenas <[email protected]>

---------

Signed-off-by: Panos Vagenas <[email protected]>

---------

Signed-off-by: Christoph Auer <[email protected]>
Signed-off-by: Panos Vagenas <[email protected]>
Signed-off-by: Christoph Auer <[email protected]>
Co-authored-by: Panos Vagenas <[email protected]>

Author: cau-git

docling_core/transforms/visualizer/layout_visualizer.py

@@ -40,6 +40,7 @@ class Params(BaseModel):
         """Layout visualization parameters."""
 
         show_label: bool = True
+        content_layers: set[ContentLayer] = {cl for cl in ContentLayer}
 
     base_visualizer: Optional[BaseVisualizer] = None
     params: Params = Params()

docling_core/transforms/visualizer/reading_order_visualizer.py

@@ -5,6 +5,7 @@
 
 from PIL import ImageDraw
 from PIL.Image import Image
+from pydantic import BaseModel
 from typing_extensions import override
 
 from docling_core.transforms.visualizer.base import BaseVisualizer
@@ -14,7 +15,16 @@
 class ReadingOrderVisualizer(BaseVisualizer):
     """Reading order visualizer."""
 
+    class Params(BaseModel):
+        """Layout visualization parameters."""
+
+        show_label: bool = True
+        content_layers: set[ContentLayer] = {
+            cl for cl in ContentLayer if cl != ContentLayer.BACKGROUND
+        }
+
     base_visualizer: Optional[BaseVisualizer] = None
+    params: Params = Params()
 
     def _draw_arrow(
         self,
@@ -71,7 +81,7 @@ def _draw_doc_reading_order(
         my_images: dict[Optional[int], Image] = images or {}
         prev_page = None
         for elem, _ in doc.iterate_items(
-            included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE},
+            included_content_layers=self.params.content_layers,
         ):
             if not isinstance(elem, DocItem):
                 continue

docling_core/types/doc/document.py

@@ -623,6 +623,7 @@ class ContentLayer(str, Enum):
 
     BODY = "body"
     FURNITURE = "furniture"
+    BACKGROUND = "background"
 
 
 DEFAULT_CONTENT_LAYERS = {ContentLayer.BODY}
@@ -860,6 +861,7 @@ class TextItem(DocItem):
         DocItemLabel.PARAGRAPH,
         DocItemLabel.REFERENCE,
         DocItemLabel.TEXT,
+        DocItemLabel.EMPTY_VALUE,
     ]
 
     orig: str  # untreated representation
@@ -2867,23 +2869,46 @@ def _with_pictures_refs(
 
     def print_element_tree(self):
         """Print_element_tree."""
-        for ix, (item, level) in enumerate(self.iterate_items(with_groups=True)):
+        for ix, (item, level) in enumerate(
+            self.iterate_items(
+                with_groups=True,
+                traverse_pictures=True,
+                included_content_layers={cl for cl in ContentLayer},
+            )
+        ):
             if isinstance(item, GroupItem):
                 print(
                     " " * level,
                     f"{ix}: {item.label.value} with name={item.name}",
                 )
+            elif isinstance(item, TextItem):
+                print(
+                    " " * level,
+                    f"{ix}: {item.label.value}: {item.text[:min(len(item.text), 100)]}",
+                )
+
             elif isinstance(item, DocItem):
                 print(" " * level, f"{ix}: {item.label.value}")
 
     def export_to_element_tree(self) -> str:
         """Export_to_element_tree."""
         texts = []
-        for ix, (item, level) in enumerate(self.iterate_items(with_groups=True)):
+        for ix, (item, level) in enumerate(
+            self.iterate_items(
+                with_groups=True,
+                traverse_pictures=True,
+                included_content_layers={cl for cl in ContentLayer},
+            )
+        ):
             if isinstance(item, GroupItem):
                 texts.append(
                     " " * level + f"{ix}: {item.label.value} with name={item.name}"
                 )
+            elif isinstance(item, TextItem):
+                texts.append(
+                    " " * level
+                    + f"{ix}: {item.label.value}: {item.text[:min(len(item.text), 100)]}"
+                )
             elif isinstance(item, DocItem):
                 texts.append(" " * level + f"{ix}: {item.label.value}")
 

docling_core/types/doc/labels.py

@@ -27,6 +27,9 @@ class DocItemLabel(str, Enum):
     KEY_VALUE_REGION = "key_value_region"
     GRADING_SCALE = "grading_scale"  # for elements in forms, questionaires representing a grading scale
     # e.g. [strongly disagree | ... | ... | strongly agree]
+    # e.g. ★★☆☆☆
+    HANDWRITTEN_TEXT = "handwritten_text"
+    EMPTY_VALUE = "empty_value"  # used for empty value fields in fillable forms
 
     # Additional labels for markup-based formats (e.g. HTML, Word)
     PARAGRAPH = "paragraph"
@@ -60,6 +63,9 @@ def get_color(label: "DocItemLabel") -> Tuple[int, int, int]:
             DocItemLabel.KEY_VALUE_REGION: (183, 65, 14),
             DocItemLabel.PARAGRAPH: (255, 255, 153),
             DocItemLabel.REFERENCE: (176, 224, 230),
+            DocItemLabel.GRADING_SCALE: (255, 204, 204),
+            DocItemLabel.HANDWRITTEN_TEXT: (204, 255, 204),
+            DocItemLabel.EMPTY_VALUE: (220, 220, 220),
         }
         return color_map.get(label, (0, 0, 0))
 
@@ -166,7 +172,6 @@ class GraphCellLabel(str, Enum):
     KEY = "key"  # used to designate a key (label) of a key-value element
     VALUE = "value"  # Data value with or without explicit Key, but filled in,
     # e.g. telephone number, address, quantity, name, date
-    EMPTY_VALUE = "empty_value"  # used for empty value fields in fillable forms
     CHECKBOX = "checkbox"
 
     def __str__(self):
@@ -179,7 +184,6 @@ def get_color(label: "GraphCellLabel") -> Tuple[int, int, int]:
         color_map = {
             GraphCellLabel.KEY: (255, 0, 0),
             GraphCellLabel.VALUE: (0, 255, 0),
-            GraphCellLabel.EMPTY_VALUE: (0, 0, 255),
         }
         return color_map.get(label, (0, 0, 0))
 

docs/DoclingDocument.json

@@ -361,7 +361,8 @@
       "description": "ContentLayer.",
       "enum": [
         "body",
-        "furniture"
+        "furniture",
+        "background"
       ],
       "title": "ContentLayer",
       "type": "string"
@@ -711,7 +712,6 @@
         "unspecified",
         "key",
         "value",
-        "empty_value",
         "checkbox"
       ],
       "title": "GraphCellLabel",
@@ -2099,7 +2099,8 @@
             "page_header",
             "paragraph",
             "reference",
-            "text"
+            "text",
+            "empty_value"
           ],
           "title": "Label",
           "type": "string"