TLDR-420 word bbox annotation (#307)

* add bbox annotations for words

* update tests

* fix lint

* fix confidence annotation and save relative bbox in BBoxAnnotation

* add docstrings for page_width and page_height

dedoc/data_structures/bbox.py

@@ -102,6 +102,16 @@ def to_dict(self) -> dict:
         res["height"] = self.height
         return res
 
+    def to_relative_dict(self, page_width: int, page_height: int) -> dict:
+        res = OrderedDict()
+        res["x_top_left"] = self.x_top_left / page_width
+        res["y_top_left"] = self.y_top_left / page_height
+        res["width"] = self.width / page_width
+        res["height"] = self.height / page_height
+        res["page_width"] = page_width
+        res["page_height"] = page_height
+        return res
+
     @staticmethod
     def from_dict(some_dict: Dict[str, int]) -> "BBox":
         return BBox(**some_dict)

dedoc/data_structures/concrete_annotations/bbox_annotation.py

@@ -8,21 +8,22 @@
 
 class BBoxAnnotation(Annotation):
     """
-    Coordinates of the line's bounding box (in pixels) - for pdf documents.
+    Coordinates of the line's bounding box (in relative coordinates) - for pdf documents.
     """
     name = "bounding box"
 
-    def __init__(self, start: int, end: int, value: BBox) -> None:
+    def __init__(self, start: int, end: int, value: BBox, page_width: int, page_height: int) -> None:
         """
         :param start: start of the annotated text (usually zero)
         :param end: end of the annotated text (usually end of the line)
         :param value: bounding box where line is located
+        :param page_width: width of original image with this bbox
+        :param page_height: height of original image with this bbox
         """
-        try:
-            BBox(value.x_top_left, value.y_top_left, value.width, value.height)
-        except ValueError:
+        if not isinstance(value, BBox):
             raise ValueError("the value of bounding box annotation should be instance of BBox")
-        super().__init__(start=start, end=end, name=BBoxAnnotation.name, value=json.dumps(value.to_dict()))
+
+        super().__init__(start=start, end=end, name=BBoxAnnotation.name, value=json.dumps(value.to_relative_dict(page_width, page_height)))
 
     @staticmethod
     def get_api_dict(api: Api) -> Model:
@@ -31,5 +32,5 @@ def get_api_dict(api: Api) -> Model:
             "end": fields.Integer(description="annotation end index", required=True, example=4),
             "value": fields.String(description="bounding box of text chunk",
                                    required=True,
-                                   example='{"x_top_left": 0, "y_top_left": 0, "width": 70, "height": 20}')
+                                   example='{"x_top_left": 0, "y_top_left": 0, "width": 0.5, "height": 0.2, "page_width": 1000, "page_height": 400}')
         })

dedoc/data_structures/concrete_annotations/confidence_annotation.py

@@ -13,14 +13,14 @@ def __init__(self, start: int, end: int, value: str) -> None:
         """
         :param start: start of the text
         :param end: end of the text (not included)
-        :param value: confidence level in "percents" (float or integer number from 0 to 100)
+        :param value: confidence level in "percents" (float number from 0 to 1)
         """
         try:
-            assert 0.0 <= float(value) <= 100.0
+            assert 0.0 <= float(value) <= 1.0
         except ValueError:
             raise ValueError("the value of confidence annotation should be float value")
         except AssertionError:
-            raise ValueError("the value of confidence annotation should be in range [0, 100]")
+            raise ValueError("the value of confidence annotation should be in range [0, 1]")
         super().__init__(start=start, end=end, name=ConfidenceAnnotation.name, value=value, is_mergeable=False)
 
     @staticmethod

dedoc/readers/pdf_reader/data_classes/text_with_bbox.py

@@ -4,7 +4,6 @@
 
 from dedoc.data_structures.annotation import Annotation
 from dedoc.data_structures.bbox import BBox
-from dedoc.data_structures.concrete_annotations.bbox_annotation import BBoxAnnotation
 from dedoc.data_structures.serializable import Serializable
 
 
@@ -24,8 +23,6 @@ def __init__(self,
         self.text = text
         self.label = label
         self.annotations = [] if annotations is None else annotations
-        if BBoxAnnotation.name not in [annotation.name for annotation in self.annotations]:
-            self.annotations.append(BBoxAnnotation(start=0, end=len(text), value=bbox))
         self.uid = f"bbox_{uuid1()}" if uid is None else uid
 
     def __str__(self) -> str:

dedoc/readers/pdf_reader/pdf_image_reader/ocr/ocr_line_extractor.py

@@ -64,7 +64,9 @@ def __split_image2bboxes(self, image: np.ndarray, page_num: int, language: str,
                 output_dict = get_text_with_bbox_from_document_page(image, language, ocr_conf_thr)
         else:
             output_dict = get_text_with_bbox_from_cells(image, language, ocr_conf_threshold=0.0)
-        line_boxes = [TextWithBBox(text=line.text, page_num=page_num, bbox=line.bbox, line_num=line_num, annotations=line.get_confidence())
+
+        height, width = image.shape[:2]
+        line_boxes = [TextWithBBox(text=line.text, page_num=page_num, bbox=line.bbox, line_num=line_num, annotations=line.get_annotations(width, height))
                       for line_num, line in enumerate(output_dict.lines)]
 
         return line_boxes

dedoc/readers/pdf_reader/pdf_image_reader/ocr/ocr_page/ocr_line.py

@@ -1,6 +1,8 @@
 from typing import List
 
+from dedoc.data_structures.annotation import Annotation
 from dedoc.data_structures.bbox import BBox
+from dedoc.data_structures.concrete_annotations.bbox_annotation import BBoxAnnotation
 from dedoc.data_structures.concrete_annotations.confidence_annotation import ConfidenceAnnotation
 from dedoc.readers.pdf_reader.pdf_image_reader.ocr.ocr_page.ocr_tuple import OcrElement
 from dedoc.readers.pdf_reader.pdf_image_reader.ocr.ocr_page.ocr_word import OcrWord
@@ -20,17 +22,20 @@ def __init__(self, order: int, bbox: BBox, words: List[OcrWord]) -> None:
     def text(self) -> str:
         return " ".join(word.text for word in self.words if word.text != "") + "\n"
 
-    def get_confidence(self) -> List[ConfidenceAnnotation]:
+    def get_annotations(self, page_width: int, page_height: int) -> List[Annotation]:
         start = 0
         annotations = []
 
         for word in self.words:
             if word.text == "":
                 continue
 
-            annotations.append(ConfidenceAnnotation(start, start + len(word.text), str(word.confidence)))
+            end = start + len(word.text)
+            annotations.append(ConfidenceAnnotation(start, end, str(word.confidence / 100)))
+            annotations.append(BBoxAnnotation(start, end, word.bbox, page_width, page_height))
             start += len(word.text) + 1
 
+        annotations.append(BBoxAnnotation(0, start, self.bbox, page_width, page_height))
         return annotations
 
     @staticmethod

tests/api_tests/test_api_format_pdf.py

@@ -1,5 +1,6 @@
 import os
 
+from dedoc.data_structures import BBoxAnnotation
 from dedoc.data_structures.concrete_annotations.bold_annotation import BoldAnnotation
 from dedoc.data_structures.concrete_annotations.confidence_annotation import ConfidenceAnnotation
 from dedoc.data_structures.concrete_annotations.spacing_annotation import SpacingAnnotation
@@ -20,6 +21,7 @@ def __check_example_file(self, result: dict) -> None:
         self.assertIn(BoldAnnotation.name, annotation_names)
         self.assertIn(SpacingAnnotation.name, annotation_names)
         self.assertIn(ConfidenceAnnotation.name, annotation_names)
+        self.assertIn(BBoxAnnotation.name, annotation_names)
         self._check_similarity("1.2.1 Поясним за непонятное", content[3]["subparagraphs"][0]["text"])
 
     def __check_metainfo(self, metainfo: dict, actual_type: str, actual_name: str) -> None: